Its book for FEM by gokhale which is not available in Scribd
Its book for FEM by gokhale which is not available in ScribdFull description
Its book for FEM by gokhale which is not available in Scribd
CDES is providing accurate and precise Finite Element Analysis at cost effective rates to worldwide clients.
FINITE ELEMENT ANALYSISDescripción completa
Its book for FEM by gokhale which is not available in Scribd
Descrição: Its book for FEM by gokhale which is not available in Scribd
Finite Element Analysis
Descrição completa
FEA University Qus BankFull description
fea
finite element analysis by seshuFull description
Full description
Full description
tl5Fb'%59i@i&d,miI9'g;.4NW+
i!%SmcWEHrtj'lf-LI!!S&SW 3
ft
,
¥tp!.ti!&£!!%IE?ii5k
i
SHuuS!
$i-'*".
Hrih45¥~
... f..,
"b
..Fi iiiU4i , JR* j!
FUN AMEN L FINITE ELE ENT ANALY IS ND APPllC I NS With Mathematica® and MATLAB® Computations
M. ASGHAR
BHATT~
~
WILEY
JOHN WILEY 8t SONS, INC.
METU LIBRARY
Mathematica is a registeredtrademarkof WolframResearch,Inc.
MATLAB is a registered trademarkof The MathWorks, Inc. ANSYS is a registered trademarkof ANSYS, Inc. ABAQUS is a registered trademarkof ABAQUS, Inc. This book is printed on acid-freepaper.
Bhatti, M. Asghar Fundamental finiteelement analysis and applications: with Mathematica and Matlab computations/ M. Asghar Bhatti. p. cm. Includes index. ISBN 0,471-64808-6 1. Structural analysis (Engineering) 2. Finite element method, J, Title. TA646.B56 2005 620' .001'51825-dc22 Printed in the United States of America 1098765432 r,~~; ~,:: Vr /\. f·f (; ".+ TL'~'}j i'J\If,!~ ~J:'''(~
CONTENTS
CONTENTS OF THE BOOK WEB SITE PREFACE 1 FINITE ELEMENT METHOD: THE BIG PICTURE 1.1 Discretization and Element Equations / 2 1.1.1 Plane Truss Element / 4 1.1.2 Triangular Element for Two-Dimensional Heat Flow / 7 1.1.3 General Remarks on Finite Element Discretization / 14 1.1.4 Triangular Element for Two-Dimensional Stress Analysis / 16 1.2 Assembly of Element Equations -/ 21 1.3 Boundary Conditions and Nodal Solution / 36 1.3.1 Essential Boundary Conditions by Rearranging Equations / 37 1.3.2 Essential Boundary Conditions by Modifying Equations / 39 1.3.3 Approximate Treatment of Essential Boundary Conditions / 40 1.3.4 Computation of Reactions to Verify Overall Equilibrium / 41 1.4 Element Solutions and Model Validity / 49 1.4.1 Plane Truss Element / 49 1.4.2 Triangular Element for Two-Dimensional Heat Flow / 51 1.4.3 Triangular Element for Two-Dimensional Stress Analysis / 54 1.5 Solution of Linear Equations / 58 1.5.1 Solution Using Choleski Decomposition / 58 1.5.2 ConjugateGradientMethod / 62
xi xiii
1
v
vi
CONTENTS
1.6
1.7
2
Multipoint Constraints / 72 1.6.1 Solution Using Lagrange Multipliers / 75 1.6.2 Solution Using Penalty Function / 79 Units / 83
MATHEMATICAL FOUNDATION OF THE FINITE ELEMENT METHOD 98 2.1 Axial Deformation of Bars / 99 2.1.1 Differential Equation for Axial Deformations I 99 2.1.2 Exact Solutions of Some Axial Deformation Problems / 101 2.2 Axial Deformation of Bars Using Galerkin Method / 104 2.2.1 Weak Form for Axial Deformations / 105 2.2.2 Uniform Bar Subjected to Linearly Varying Axial Load / 109 2.2.3 Tapered Bar Subjected to Linearly Varying Axial Load / 113 2.3 One-Dimensional BVJ;>Using .Galerkin Method / 115 _...• -,2.3.1 Overall Solution Procedure Using GalerkinMethod / 115 2.3.2 Highet Order Boundary Value Problems / 119 2.4 Rayleigh-Ritz Method / 128 2.4.1 Potential Energy for Axial Deformation of Bars / 129 2.4.2 Overall Solution Procedure Using the Rayleigh-Ritz Method / 130 2.4.3 Uniform Bar Subjected to Linearly Varying Axial Load I 131 2.4.4 Tapered Bar Subjected to Linearly Varying Axial Load / 133 2.5 Comments on Galerkin and Rayleigh-Ritz Methods / 135 2.5.1 Admissible Assumed S~lution / 135 2.5.2 Solution Convergence-the Completeness Requirement / 136 2.5.3 Galerkin versus Rayleigh-Ritz / 138 2.6 Finite Element Form of Assumed Solutions / 138 2.6.1 LinearInterpolation Functions for Second-Order Problems / 139 2.6.2 Lagrange Interpolation / 142 2.6.3 Galerkin Weighting Functions in Finite Element Form / 143 2.9.4 Hermite Interpolation for Fourth-Order Problems / 144 2.7 Finite Element Solution of Axial Deformation Problems / 150 2.7.1 Two-Node Uniform Bar Element for Axial Deformations / 150 2.7.2 Numerical Examples / 155
3 ONE-DIMENSIONAL BOUNDARY VALUE PROBLEM 3.1 Selected Applications of 1D BVP / 174 3.1.1 Steady-State Heat Conduction / 174 3.1.2 Heat Flow through Thin Fins / 175
173
CONTENTS
3.1.3 Viscous Fluid Flow between Parallel Plates-Lubrication Problem / 176 3.1.4 Slider Bearing / 177 3.1.5 Axial Deformation of Bars / 178 3.1.6 Elastic Buckling of Long Slender Bars / 178 3.2 Finite Element Formulation for Second-Order ID BVP / 180 3.2.1 Complete Solution Procedure / 186 3.3 Steady-State Heat Conduction / 188 3.4 Steady-State Heat Conduction and Convection / 190 3.5 Viscous Fluid Flow Between Parallel Plates / 198 3.6 Elastic Buckling of Bars / 202 3.7 Solution of Second-Order 1D BVP / 208 3.8 A Closer Look at the Interelement Derivative Terms / 214 4 TRUSSES, BEAMS, AND FRAMES
4.1 Plane Trusses / 223 4.2 Space Trusses / 227 4.3 Temperature Changes and Initial Strains in Trusses / 231 4.4 Spring Elements / 233 4.5 Transverse Deformation of Beams / 236 4.5.1 Differential Equation for Beam Bending / 236 4.5.2 Boundary Conditions for Beams / 238 4.5.3 Shear Stressesin Beams / 240 4.5.4 Potential Energy for Beam Bending / 240 4.5.5 Transverse Deformation of a Uniform Beam / 241 4.5.6 Transverse Deformation of a Tapered Beam Fixed at Both Ends / 242 4.6 Two-Node Beam Element / 244 4.6.1 Cubic Assumed Solution / 245 4.6.2 Element Equations Using Rayleigh-Ritz Method / 246 4.7 Uniform Beams Subjected to Distributed Loads / 259 4.8 Plane Frames / 266 4.9 Space Frames / 279 4.9.1 Element Equations in Local Coordinate System / 281 4.9.2 Local-to-Global Transformation / 285 4.9.3 Element Solution / 289 4.10 Frames in Multistory Buildings / 293
222
vii
viii
CONTENTS
5 TWO-DIMENSIONALELEMENTS 5.1 Selected Applications of the 2D BVP / 313 5.1.1 Two-Dimensional Potential Flow / 313 5.1.2 Steady-State Heat Flow / 316 5.1.3 Bars Subjected to Torsion / 317 5.1.4 Waveguidesin Electromagnetics / 319 5.2 Integration by Parts in Higher Dimensions / 320 5.3 Finite Element Equations Using the Galerkin Method / 325 5.4 Rectangular Finite Elements / 329 5.4.1 Four-Node Rectangular Element / 329 5.4.2 Eight-Node Rectangular Element / 346 5.4.3 Lagrange Interpolation for Rectangular Elements / 350 5.5 Triangular Finite Elements / 357 5.5.1 Three-Node Triangular Element / 358 5.5.2 Higher Order Triangular Elements / 371
311
6
381
MAPPED ELEMENTS 6) Integration Using Change of Variables / 382 6.1.1 One-Dimensional Integrals / 382 6.1.2 Two-Dimensional Area Integrals / 383 6.1.3 Three-Dimensional VolumeIntegrals / 386 6.2 Mapping Quadrilaterals Using Interpolation Functions / 387 6.2.1 Mapping Lines / 387 6.2.2 Mapping Quadrilater~ Areas / 392 6.2.3 Mapped Mesh Gene~ation / 405 6.3 Numerical Integration Using Gauss Quadrature / 408 6.3.1 Gauss Quadrature for One-Dimensional Integrals / 409 6.3.2 Gauss Quadrature for Area Integrals / 414 6.3.3 Gauss Quadrature for VolumeIntegrals / 417 6.4 Finite Element Computations Involving Mapped Elements / 420 6.4.1 Assumed Solution / 421 6.4.2 Derivatives of the Assumed Solution / 422 6.4.3 Evaluation of Area Integrals / 428 6.4.4 Evaluation of Boundary Integrals / 436 6.5 Complete Mathematica and MATLAB Solutions of 2D BVP Involving Mapped Elements / 441 6.6 Triangular Elements by Collapsing Quadrilaterals / 451 6.7 Infinite Elements / 452 6.7.1 One-DirnensionalBVP / 452 6.7.2 Two-Dimensional BVP / 458
CONTENTS
7 ANALYSIS OF ELASTIC SOLIDS 467 7.1 Fundamental Concepts in Elasticity / 467 7.1.1 Stresses / 467 7.1.2 Stress Failure Criteria / 472 7.1.3 Strains / 475 7.1.4 Constitutive Equations / 478 7.1.5 TemperatureEffects and Initial Strains / 480 7.2 GoverningDifferential Equations / 480 7.2.1 Stress Equilibrium Equations / 481 7.2.2 Governing Differential Equations in Terms of Displacements / 482 7.3 General Form of Finite Element Equations / 484 7.3.1 Potential Energy Functional / 484 7.3.2 Weak Form / 485 7.3.3 Finite Element Equations / 486 7.3,4 Finite Element Equations in the Presence of Initial Strains / 489 7.4 Plane Stress and Plane Strain / 490 7.4.1 Plane Stress Problem / 492 7.4.2 Plane Strain Problem / 493 7.4.3 Finite Element Equations / 495 7.4.4 Three-Node Triangular Element / 497 7.4.5 Mapped Quadrilateral Elements / 508 7.5 Planar Finite Element Models / 517 7.5.1 Pressure Vessels / 517 7.5.2 Rotating Disks and Flywheels / 524 7.5.3 Residual Stresses Due to Welding / 530 7.5.4 Crack Tip Singularity / 531 8 TRANSIENT PROBLEMS 8.1 TransientField Problems / ,545 8.1.1 Finite Element Equations / 546 8.1.2 Triangular Element / 549 8.1.3 Transient Heat Flow / 551 8.2 Elastic Solids Subjected to Dynamic Loads / 557 8.2.1 Finite Element Equations / 559 8.2.2 Mass Matrices for Common Structural Elements / 561 8.2.3 Free-VibrationAnalysis / 567 8.2.4 Transient Response Examples / 573
545
lx
x
CONTENTS
9 p-FORMULATION· 9.1 p-Formulation for Second-Order 1D BVP / 586 9.1.1 Assumed Solution Using Legendre Polynomials / 587 9.1.2 Element Equations / 591 9.1.3 Numerical Examples / 593 9.2 p-Formulation for Second-Order 2D BVP / 604 9.2.1 p-Mode Assumed Solution / 605 9.2.2 Finite Element Equations / 608 9.2.3 Assembly of Element Equations / 617 9.2.4 Incorporating Essential Boundary Conditions / 620 9.2.5 Applications / 624
B VARIATIONAL FORM FOR BOUNDARY VALUE PROBLEMS B.1 Basic Concept of Variation of a Function / 676 B.2 Derivation of Equivalent Variational Form / 679 B.3 Boundary Value Problem Corresponding to a Given Functional / 683
676
BIBLIOGRAPHY
687
INDEX
695
CONTEN'TS OF THE BOOK WEB S~TE (www.wiley.com/go/bhatti)
ABAQUS Applications AbaqusUse\AbaqusExecutionProcedure.pdf Abaqus Use\HeatFlow AbaqusUse\PlaneStress Abaqus Use\TmssAnalysis ANSYS Applications AnsysUse\AppendixA . AnsysUse\Chap5 AnsysUse\Chap7 AnsysUse\Chap8 AnsysUse\GeneralProcedure.pdf Full Detail Text Examples Full Detail Text Examples\ChaplExarnples.pdf Full Detail Text Examples\Chap2Examples.pdf Full Detail Text Examples\Chap3Examples.pdf . Full Detail Text Exarnples\Chap4Exarnples.pdf Full Detail Text Exarnples\Chap5Exarnples.pdf Full Detail Text Examples\Chap6Examples.pdf Full Detail Text Examples\Chap7Examples.pdf xi
xii
CONTENTS OF THE BOOKWEB SITE
Full Detail Text Examples\Chap8Examples.pdf Full Detail Text Examples\Chap9Examples.pdf Mathematica Applications MathematicaUse\MathChap l.nb MathematicaUse\MathChap2.nb MathematicaUse\MathChap3.nb MathematicaUse\MathChap4.nb MathematicaUse\MathChap5 .nb MathematicaUse\MathChap6.nb MathematicaUse\MathChap7.nb MathematicaUse\MathChap8.nb MathematicaUse\Mathematica Introduction.nb MATLAB Applications MatlabFiles\Chap I MatlabFiles\Chap2 MatlabFiles\Chap3 MatlabFiles\Chap4 MatlabFiles\Chap5 MatlabFiles\Chap6 MatlabFiles\Chap7 MatlabFiles\Chap8 MatlabFiles\Common
I
Sample Course Outlines, Lectures, and Examinations Supplementary Material and Corrections
PREFACE
Large numbers of books have been written on the finite element method. However, effective teaching of the method using most existing books is a difficult task. The vast majority of current books present the finite element method as an extension of the conventional matrix structural analysis methods. Using this approach, one can teach the mechanical aspects of the finite element method fairly well, but there are no satisfactory explanations for even the simplest theoretical questions. Why are rotational degrees of freedom defined for the beam and plate elements but not for the plane stress and truss elements? What is wrong with connecting corner nodes of a planar four-node element to the rnidside nodes of an eight-node element? The application of the method to nonstructural problems is possible only if one can interpret problem parameters in terms of their structural counterparts. For example, one can solve heat transfer problems because temperature can be interpreted as displacement in a structural problem. More recently, several new textbooks on finite elements have appeared that emphasize the mathematical basis of the finite element method. Using some of these books, the finite element method can be presented as a method for .obtaining approximate solution of ordinary and partial differential equations. The choice of appropriate degrees of freedom, boundary conditions, trial solutions, etc., can now be fully explained with this theoretical background. However, the vast majority of these books tend to be too theoretical and do not present enough computational details and examples to be of value, especially to undergraduate and first-year graduate students in engineering. The finite element coursesface one more hurdle. One needs to perform computations in order to effectively learn the finite element techniques. However, typical finite element calculations are very long and tedious, especially those involving mapped elements. In fact, some of these calculations are essentially impossible to perform by hand. To alleviate this situation, instructors generally rely on programs written in FORTRAN or some other xiii
xiv
PREFACE
conventional programming language. In fact, there are several books available that include these types of programs with them. However, realistically, in a typical one-semester course, most students cannot be expected to fully understand these programs. At best they use them as black boxes, which obviously does not help in learning the concepts. In addition to traditional research-oriented students, effective finite element courses must also cater to the needs and expectations of practicing engineers and others interested only in the finite element applications. Knowing the theoretical details alone does not help in creating appropriate models for practical, and often complex, engineering systems. This book is intended to strike an appropriate balance among the theory, generality, and practical applications of the finite element method. The method is presented as a fairly straightforward extension of the classical weighted residual and the Rayleigh-Ritz methods for approximate solution of differential equations. The theoretical details are presented in an informal style appealing to the reader's intuition rather than mathematical rigor. To make the concepts clear, all computational details are fully explained and numerous examples are included showing all calculations. To overcome the tedious nature of calculations associated with finite elements, extensive use of MATLAB® and Mathematicd'' is made in the boole. All finite element procedures are implemented in the form of interactive Mathematica notebooks and easy-to-follow MATLAB code. All necessary computations are readily apparent from these implementations. Finally, to address the practical applications of the finite element method, the book integrates a series of computer laboratories and projects that involve modeling and solution using commercial finite element software. Short tutorials and carefully chosen sample applications of ANSYS and ABAQUS are contained in the book. The book is organized in such a way that it can be used very effectively in a lecture/ computer laboratory (lab) format. In over 20 years of teaching finite elements, using a variety of approaches, the author has found that presenting the material in a two-hour lecture and one-hour lab per week is i~eally suited for the first finite element course. The lecture part develops suitable theoretical background while the lab portion gives students experience in finite element modeling and actual applications. Both parts should be taught in parallel. Of course, it takes time to develop the appropriate theoretical background in the lecture part. The lab part, therefore, is ahead of the lectures and, in the initial stages, students are using the finite element software essentially as a black box. However, this approach has two main advantages. The first is that students have some time to get familiar with the particular computer system and the finite element package being utilized. The second, and more significant, advantage is that it raises students' curiosity in learning more about why things must be done in a certain way. During early labs students often encounter errors such as "negative pivot found" or "zero or negative Jacobian for element." When, during the lecture part, they find out mathematical reasons for such errors, it makes them appreciate the importance of learning theory in order to become better users of the finite element technology. The author also feels strongly that the labs must utilize one of the several commercially available packages, instead of relying on simple home-grown programs. Use of commercial programs exposes students to at least one state-of-the-art finite element package with its built-in or associated pre- and postprocessors. Since the general procedures are very similar among different programs, it is relatively easy to learn a different package after this
PREFACE
exposure. Most commercial prol$nims also include analysis modules for linear and nonlinear static and dynamic analysis, buclding, fluid flow, optimization, and fatigue. Thus with these packages students can be exposed to a variety of finite element applications, even though there generally is not enough time to develop theoretical details of all these topics in one finite element course. With more applications, students also perceive the course as more practical and seem to put more effort into learning.
TOPICS COVERED The book covers the fundamental concepts and is designed for a first course on finite elements suitable for upper division undergraduate students and first-year graduate students. It presents the finite element method as a tool to find approximate solution of differential equations and thus can be used by students from a variety of disciplines. Applications covered include heat flow, stress analysis, fluid flow, and analysis of structural frameworks. The material is presented in nine chapters and two appendixes as follows.
1. Finite Element Method: The Big Picture. This chapter presents an overview of the finite element method. To give a clear idea of the solution process, the finite element equations for a few simple elements (plane truss, heat flow, and plane stress) are presented in this chapter. A few general remarks on modeling and discretization are also included. Important steps of assembly, handling boundary conditions, and solutions for nodal unknowns and element quantities are explained in detail in this chapter. These steps are fairly mechanical in nature and do not require complex theoretical development. They are, however, central to actually obtaining a finite element solution for a given problem. The chapter includes brief descriptions of both direct and iterative methods for solution oflinear systems of equations. Treatment of linear constraints through Lagrange multipliers and penalty functions is also included. This chapter gives enough background to students so that they can quickly start using available commercial finite element packages effectively. It plays an important role in the lecture/lab format advocated-for the first finite element course. 2. Mathematical Foundations of the Finite Element Method. From a mathematical point of view the finite element method is a special form of the well-known Galerkin and Rayleigh-Ritz methods for finding approximate solutions of differential equations. The basic concepts are explained in this chapter with reference to the problem of axial deformation of bars. The derivation of the governing differential equation is included for completeness. Approximate solutions using the classical form of Galerkin and RayleighRitz methods are presented. Finally, the methods are cast into the form that is suitable for developing finite element equations. Lagrange and Hermitian interpolation functions, commonly employed in derivation of finite element equations, are presented in this chapter. 3. One-Dimensional Boundary Value Problem. A large humber of practical problems are governed by a one-dimensional boundary value problem of the form d (
dU(X))
dx k(x)~ + p(x) u(x) + q(x)
=0
xv
xvi
PREFACE
Finite element formulation and solutions of selected applications that are governed by the differential equation of this form are presented in this chapter. 4. Trusses, Beams, and Frames. Many structural systems used in practice consist of long slender members of various shapes used in trusses, beams, and frames. This chapter presents finite element equations for these elements. The chapter is important for civil and mechanical engineering students interested in structures. It also covers typical modeling techniques employed in framed structures, such as rigid end zones and rigid floor diaphragms. Those not interested in these applications can skip this chapter without any loss in continuity. 5. Two-Dimensional Elements. In this chapter the basic finite element concepts are il-lustrated with reference to the following partial differential equation defined over an arbitrary two-dimensional region:
The equation can easily be recognized as a generalization of the one-dimensional boundary value problem considered in Chapter 3. Steady-state heat flow, a variety of fluid flow, and the torsion of planar sections are some of the common engineering applications that are governed by the differential equations that are special cases. of this general boundary value problem. Solutions of these problems using rectangular and triangular elements are presented in this chapter. 6. Mapped Elements. Quadrilateral elements and other elements that can have curved sides are much more useful in accurately modeling arbitrary shapes. Successful development of these elements is based on the key concept of mapping. These concepts are discussed in this chapter. Derivation of the Gaussian quadrature used to evaluate equations for mapped elements is presented. Four-sand eight-node quadrilateral elements are presented for solution of two-dimensional boundary value problems. The chapter also includes procedures for forming triangles by collapsing quadrilaterals and for developing the so-called infinite elements to handle far-field boundary conditions. 7. Analysis ofElastic Solids. The problem of determining stresses and strains in elastic solids subjected to loading and temperature changes is considered in this chapter. The fundamental concepts from elasticity are reviewed. Using these concepts, the governing differential equations in terms of stresses and displacements are derived followed by the general form of finite element equations for analysis of elastic solids. Specific elements for analysis of plane stress and plane strain problems are presented in this chapter. The so-called singularity elements, designed to capture a singular stress field near a crack tip, are discussed. This chapter is important for those interested in stress analysis. Those not interested in these applications can skip this chapter without any loss in continuity. 8.· Transient Problems. This chapter considers analysis of transient problems using finite elements. Formulations for both the transient field problems and the structural dynamics problems are presented in this chapter. 9. p-Formulation. In conventional finite element formulation, each element is based on a specific set of interpolation functions. After choosing an element type, the only way to
PREFACE
obtain a better solution is to refihe the model. This formulation is called h-formulation, where h indicates the generic size of an element. An alternative formulation, called the p-formulation, is presented in this chapter. In this formulation, the elements are based on interpolation functions that may involve very high order terms. The initial finite element model is fairly coarse and is based primarily on geometric considerations. Refined solutions are obtained by increasing the order of the interpolation functions used in the formulation. Efficient interpolation functions have been developed so that higher order solutions can be obtained in a hierarchical manner from the lower order solutions. 10. Appendix A: Use of Commercial FEA Software. This appendix introduces students to two commonly used commercial finite element programs, ANSYS and ABAQUS. Concise instructions for solution of structural frameworks, heat flow, and stress analysis problems are given for both programs. 11. Appendix B: Variational Form for Boundary Value Problems. The main body of the text employs the Galerkin approach for solution of general boundary value problems and the variational approach (using potential energy) for structural problems. The derivation of the variational functional requires familiarity with the calculus of variations. In the author's experience, given that only limited time is available, most undergraduate students have difficulty fully comprehending this topic. For this reason, and since the derivation is not central to the finite element development, the material on developing variational functionals is moved to this appendix. If desired, this material can be covered with the discussion of the Rayleigh-Ritz method in Chapter 2.
To keep the book to a reasonable length and to make it suitable for a wider audience, important structural oriented topics, such as axisymmetric and three-dimensional elasticity, plates and shells, material and geometric nonlinearity, mixed and hybrid formulations, and contact problems are not covered in this book. These topics are covered in detail in a companion textbook by the author entitled Advanced Topics in Finite Element Analysis of Structures: With Mathematico'" and lvIATLAB® Computations, John Wiley, 2006.
UNIQUE FEATURES
(i) All key. ideas are introduced in chapters that emphasize the method as a way to find approximate solution of boundary value problems. Thus the book can be used effectively for students from a variety of disciplines.. (ii) The "big picture" chapter gives readers an overview of all the mechanical details of the finite element method very quickly. This enables instructors to start using commercial finite element software early in the semester; thus allowing plenty of opportunity to bring practical modeling issues into the classroom. The author is not aware of any other book that starts out in this manner. Few books that actually try to do this do so by taldng discrete spring and bar elements. In my experience this does not work very well because students do not see actual finite element applications. Also, this approach does not make sense to those who are not interested in structural applications.
xvll
xviii
PREFACE
(iii) Chapters 2 and 3 introduce fundamental finite element concepts through onedimensional examples. The axial deformation problem is used for a gentle introduction to the subject. This allows for parameters to be interpreted in physical terms. The derivation of the governing equations and simple techniques for obtaining exact solutions are included to help those who may not be familiar with the structural terminology. Chapter 3 also includes solution of one-dimensional boundary value problems without reference to any physical application for nonstructural readers. (iv) Chapter 4, on structural frameworks, is quite unique for books on finite elements. No current textbook that approaches finite elements from a differential equation point of view also has a complete coverage of structural frames, especially in three dimensions. In fact, even most books specifically devoted to structural analysis do not have as satisfactory a coverage of the subject as provided in this chapter. (v) Chapters 5 and 6 are two important chapters that introduce key finite element concepts in the context of two-dimensional boundary value problems. To keep the integration and differentiation issues from clouding the basic ideas, Chapter 5 starts with rectangular elements and presents complete examples using such elements. The triangular elements are presented next. By the time the mapped elements are presented in Chapter 6, there are no real finite element-related concepts left. It is all just calculus. This clear distinction between the fundamental concepts and calculus-related issues gives instructors flexibility in presenting the material to students with a wide variety of mathematics background. (vi) Chapter 9, on p-formulation, is unique. No other book geared toward the first finite element course even mentions this important formulation. Several ideas presented in this chapter are used in recent development of the so-called mesh less methods. (vii) Mathematica and MATLAB ,implementations are included to show how calculations can be organized using' a computer algebra system. These implementations require only the very basic understanding of these systems. Detailed examples are presented in Chapter 1 showing how to generate and assemble element equations, reorganize matrices to account for boundary conditions, and then solve for primary and secondary unknowns. These steps remain exactly the same for all implementations. Most of the other implementations are nothing more than element matrices written using Mathematica or MATLAB syntax. (viii) Numerous numerical examples are included to clearly show all computations involved. (ix) All chapters contain problems for homework assignment. Most chapters also contain problems suitable for computer labs and projects. The accompanying web site (www.wiley.com/go/bhatti) contains all text examples, MATLAB and Mathematica functions, and ANSYS and ABAQUS files in electronic form. To keep the printed book to a reasonable length most examples skip some computations. The web site contains full computational details of-these examples. Also the book generally alternates between showing examples done with Mathematica and MATLAB. The web site contains implementations of all examples in both Mathematica and MATLAB.
PREFACE
tVPICAL COURSES The book can be used to develop a number of courses suitable for different audiences. First Finite Element Course for Engineering Students About 32 hours of lectures and 12 hours of labs (selected materials from indicated chapters):
Chapter l: Finite element procedure, discretization, element equations, assembly, boundary conditions, solution of primary unknowns and element quantities, reactions, solution validity (4 hr) Chapter 2: Weak form for approximate solution of differential equations, Galerkin method, approximate solutions using Rayleigh-Ritz method, comparison of Galerkin and Rayleigh-Ritz methods, Lagrange and Hermite interpolation, axial deformation element using Rayleigh-Ritz and Galerkin methods (6 hr) Chapter 3: ID BVP, FEA solution ofBVP, ID BVP applications (3 hr) Chapter 4: Finite element for beam bending, beam applications, structural frames (3 hr) Chapter 5: Finite elements for 2D and 3D problems, linear triangular element for second-order 2D BVP, 2D fluid flow and torsion problems (4 hr) Chapter 6: 2D Lagrange and serendipity shape functions, mapped elements, evaluation of area integrals for 2D mapped elements, evaluation of line integrals for 2D mapped elements (4 hr) Chapter 7: Stresses and strains in solids, finite element analysis of elastic solids, CST and isoparametric elements for plane elasticity (4 hr) Chapter 8: Transient problems (2 hr) Review, exams (2 hr) About 12 hours of labs (some sections from the indicated chapters supplemented by documentation of the chosen commercial software): Appendix: Introduction to Mathematica and/or MATLAB (2 hr) Chapters 1 and 4: Software documentation, basic finite element procedure using commercial software, truss and frame problems (2 hr) Chapters 1 and 5: Software documentation, 2D mesh generation, heat flow problems (2 hr) Chapters 1 and 7: 2D, axisymmetric, and 3D stress analysis problems (2 hr) Chapter 8: Transient problems (2 hr) Software documentation: Constraints, design optimization (2 hr) First Finite Element Course for Students Not Interested in Structural Applications Skip Chapters 4 and 7. Spend more time on applications in Chapters 5 and 6. Introduce Chapter 9: p-Formulation. In the labs replace truss, frame, and stress analysis problems with appropriate applications. Finite Element Course for Practicing Engineers From the current book: Chapters 1, 2, 6, and 7. From the companion advanced book: Chapters 1, 2, and 5 and selected material from Chapters 6, 7, and 8.
xix
xx
PREFACE
Finite Element Modeling and Applications For a short course on finite element modeling or self-study, it is suggested to cover the first chapter in detail and then move on to Appendix A for specific examples of using commercial finite element packages for solution of practical problems.
ACKNOWLEDGMENTS
Most of the material presented in the book has become part of the standard finite element literature, and hence it is difficult to acknowledge contributions of specific individuals. I am indebted to the pioneers in the field and the authors of all existing books and journal papers on the subject. I have obviously benefited from their contributions and have used a good number of them in my over 20 years of teaching the subject. I wrote the first draft of the book in early 1990. However, the printed version has practically nothing in common with that first draft. Primarily as a result of questions from my students, I have had to make extensive revisions almost every year. Over the last couple of years the process began to show signs of convergence andthe result is what you see now. Thus I would like to acknowledge all direct and indirect contributions of my former students. Their questions hopefully led me to explain things in ways that make sense to most readers. (A note to future students and readers: Please keep the questions coming.) I want to thank my former graduate student Ryan Vignes, who read through several drafts of the book and provided valuable feedback. Professors Jia Liu and Xiao Shaoping used early versions of the book when they taught finite elements. Their suggestions have helped a great deat in improving the book. My colleagues Professors Ray P.S. Han, Hosin David Lee, and Ralph Stephens have helped by sharing their teaching philosophy and by keeping me in shape through heated games of badminton and tennis. Finally, I would like to acknowledge the editorial staff of John Wiley for doing a great job in the production of the book. 1'/am especially indebted to Jim Harper, who, from our first meeting in Seattle in 2003, has been in constant communication and has kept the process going smoothly. Contributions of senior production editor Bob Hilbert and editorial assistant Naomi Rothwell are gratefully acknowledged.
CHAPTER ONE 5·
FINITE ELEMENT METHOD: THE BIG PICTURE
Application of physical principles, such as mass balance, energy conservation, and equilibrium, naturally leads many engineering analysis situations into differential equations. Methods have been developed for obtaining exact solutions for various classes of differential equations. However, these methods do not apply to many practical problems because either their governing differential equations do not fall into these classes or they involve complex geometries. Finding analytical solutions that also satisfy boundary conditions specified over arbitrary two- and three-dimensional regions becomes a very difficult task. Numerical methods are therefore widely used for solution of practical problems in all branches of engineering. The finite element method is one of the numerical methods for obtaining approximate solution of ordinary and partial differential equations. It is especially powerful when dealing with boundary conditions defined over complex geometries that are common in practical applications. Other numerical methods such as finite difference and boundary element methods may be competitive or even superior to the finite element method for certain classes of problems. However, because of its versatility in handling arbitrary domains and availability of sophisticated commercial finite element software, over the last few decades, the finite element method has become the preferred method for solution of many practical problems. Only the finite element method is considered in detail in this book. Readers interested in other methods should consult appropriate references, Books by Zienkiewicz and Morgan [45], Celia and Gray [32], and Lapidus and Pinder [37] are particularly useful for those interested in a comparison of different methods. The application of the finite element method to a given problem involves the following six steps:
2
FINITEELEMENTMETHOD:THE BIG PICTURE
1. 2. 3. 4. 5. 6.
Development of element equations Discretization of solution domain into a finite element mesh Assembly of element equations Introduction of boundary conditions Solution for nodal unknowns Computation of solution and related quantities over each element
The key idea of the finite element method is to discretize the solution domain into a number of simpler domains called elements. An approximate solution is assumed over an element in terms of solutions at selected points called nodes. To give a clear idea of the overall finite element solution process, the finite element equations for a few simple elements are presented in Section 1.1. Obviously at this stage it is not possible to give derivations of these equations. The derivations must wait until later chapters after we have developed enough theoretical background. Few general remarks on discretization are also made in Section 1.1. More specific comments on modeling are presented in later chapters when discussing various applications. Important steps of assembly, handling boundary conditions, and solutions for nodal unknowns and element quantities remain essentially unchanged for any finite element analysis. Thus these procedures are explained in detail in Sections 1.2, 1.3, and 104. These steps are fairly mechanical in nature and do not require complex theoretical development. They are, however, central to actually obtaining a finite element solution for a given problem. Therefore, it is important to fully master these steps before proceeding to the remaining chapters in the book. The finite element process results in a large system of equations that must be solved for determining nodal unknowns. Several methods are available for efficient solution of these large and relatively sparse systems of equations. A brief introduction to two commonly employed methods is given in Section 1.5. In some finite element modeling situations it becomes necessary to introduce constraints in the finite element equations. Section 1.6 presents examples of few such situations and discusses two different methods for handling these so-called multipoint constraints. A brief section on appropriate use of units in numerical calculations concludes this chapter.
1.1
DISCRETIZATION AND ELEMENT EQUATIONS
Each analysis situation that is described in terms of one or more differential equations requires an appropriate set of element equations. Even for the same system of governing equations, several elements with different shapes and characteristics may be available. It is crucial to choose an appropriate element type for the application being considered. A proper choice requires knowledge of all details of element formulation and a thorough understanding of approximations introduced during its development. A key step in the derivation of element equations is an assumption regarding the solution of the goveming differential equation over an element. Several practical elements are available that assume a simple linear solution. Other elements use more sophisticated functions to describe solution over elements. The assumed element solutions are written in terms of unknown solutions at selected points called nodes. The unknown solutions at the nodes are
DISCRETIZATION AND ELEMENT EQUATIONS
generally referred to as the nodal degrees offreedom, a terminology that dates back to the early development of the method by structural engineers. The appropriate choice of nodal degrees of freedom depends on the governing differential equation and will be discussed in the following chapters. The geometry of an element depends on the type of the governing differential equation. For problems defined by one-dimensional ordinary differential equations, the elements are straight or curved line elements. For problems governed by two-dimensional partial differential equations the elements are usually of triangular or quadrilateral shape. The element sides may be straight or curved. Elements with curved sides are useful for accurately modeling complex geometries common in applications such as shell structures and automobile bodies. Three-dimensional problems require tetrahedral or solid brick-shaped elements. Typical element shapes for one-, two-, andthree-dimensional (lD, 2D, and 3D) problems are shown in Figure 1.1. The nodes on the elements are shown as dark circles. Element equations express a relationship between the physical parameters in the governing differential equations and the nodal degrees of freedom. Since the number of equations for some of the elements can be very large, the element equations are almost always written using a matrix notation. The computations are organized in two phases. In the first phase (the element derivation phase), the element matrices are developed for a typical element that is representative of all elements in the problem. Computations are performed in a symbolic form without using actual numerical values for a specific element. The goal is to develop general formulas for element matrices that can later be used for solution of any numerical problem belonging to that class. In the second phase, the general formulas are used to write specific numerical matrices for each element. One of the main reasons for the popularity of the finite element method is the wide availability of general-purpose finite element analysis software. This software development is possible because general element equations can be programmed in such a way that, given nodal coordinates and other physical parameters for an element, the program returns numerical equations for that element. Commercial finite element programs contain a large library of elements suitable for solution of a wide variety of practical problems.
ID Elements
2D Elements
3D Elements
Figure 1.1. Typical finite element shapes
3
4
FINITE ELEMENTMETHOD:THE BIG PICTURE
To give a clear picture of the overall finite element solution procedure, the general finite element equations for few commonly used elements are given below. The detailed derivations of these equations are presented in later chapters.
1.1.1
Plane Truss Element
Many structural systems used in practice consist of long slender shapes of various cross sections. Systems in which the shapes are arranged so that each member primarily resists axial forces are usually known as trusses. Common examples are roof trusses, bridge supports, crane booms, and antenna towers. Figure 1.2 shows a transmission tower that can be modeled effectively as a plane truss. For modeling purposes all members are considered pin jointed. The loads are applied at the joints. The analysis problem is to find joint displacements, axial forces, and axial stresses in different members of the truss." Clearly the basic element to analyze any plane truss structure is a two-node straightline element oriented arbitrarily in a two-dimensional x-y plane, as shown in the Figure 1.3. The element end nodal coordinates are indicated by (Xl' YI) and (x2' Y2). The element axis s runs from the first node of the element to the second node. The angle a defines the orientation of the element with respect to a global x-y coordinate system. Each node has two displacement degrees of freedom, u indicating displacement in the X direction and v indicating displacement in the y direction. The element can be subjected to loads only at its ends. Using these elements, the finite element model of the transmission tower is as shown in Figure 1.4. The model consists of 16 nodes and 29 plane truss elements. The element numbers and node numbers are assigned arbitrarily for identification purposes.
600 570 540 480 420
10001b 10001b
300
180
o 300
180
96
o
6096
180
Figure 1.2. Transmission tower
300
in
DISCRETIZATION AND ELEMENT EQUATIONS
y Nodal dof
End loads
x Figure 1.3. Plane truss element
Element numbers
Figure 1.4. Planetruss element model of the transmission tower
Using procedures discussed in later chapters, it can be shown that the finite element equations for a plane truss-dement are as follows: Is lns
-1;
In; -Is Ins -ls lns z2s I s lns -In; where E = elastic modulus of the material (Young's modulus), A = area of cross section of the element, L = length of the element, and Is. Ins are the direction cosines of the element axis (line from element node 1 to 2). Here, Is is the cosine of angle a between the element axis and the x axis (measured 'counterclockwise) and Ins is the cosine of angle between the element axis and the y axis. In terms of element nodal coordinates,
5
6
FINITE ELEMENTMETHOD:THE BIG PICTURE
In the element equations the left-hand-side coefficient matrix is usually called the stiffness matrix and the right-hand-side vector as the nodal load vector. Note that once the element end coordinates, material property, cross-sectional area, and element loading are specified, the only unknowns in the element equations are the nodal displacements. It is important to recognize that the element equations refer to an isolated element, We cannot solve for the nodal degrees of freedom for the entire structure by simply solving the equations for one element. We must consider contributions of all elements, loads, and support conditions before solving for the nodal unknowns. These procedures are discussed in detail in later sections of this chapter. Example 1.1 Write finite element equations for element number 14 in the finite element model of the transmission tower shown in Figure 1.4. The tower is made of steel (!i..=29 x 106Ib/in2 ) angle sections. The area of cross section of element 14 is 1.73 in2 . The element is connected between nodes 7 and 9. We can choose e~as th~ first node of the element. Choosing node 7 as the first node establishes the element s axis as going from node 7 toward 9. The origin of the global x-y coordinate system can be placed at any convenient location. Choosing the centerline of the tower as the origin, the nodal coordinates for the element 14 are as follows:
First node (node 7) = (-60, 420) in; Second node (node 9) = (-180, 480) in;
XI
= -60;
YI
= 420
x2
= -180;
Y2
= 480
Using these coordinates, the element length and the direction cosines can easily be calculated as follows: Element length:
L = ~(X2 -xll + (Y2
Element direction cosines:
;' _ x2 - XI _ 2. Is - - L - - - -{5'
-------
= 60-{5 in 112 = Y2 - YI = _l_
- YI)2
s
L
-{5
From the given material and section properties, E
E: =
= 29000000Ib/in2 ;
373945. lb/in
Using these values, the element stiffness matrix (the left-hand side of the element equations) can easily be written as follows:
The right-hand-side vector of element equations represents applied loads at the element ends. There are no loads applied at node 7. The applied load of 1000 lb at node 9 is shared by elements 14, 16,23, and 24. The portion taken by element 14 cannot be determined
DISCRETIZATION AND ELEMENT EQUATIONS
without detailed analysis of the tower, which is exactly what we are attempting to do in the first place. Fortunately, to proceed with the analysis, it is not necessary to know the portion of the load resisted by different elements meeting at a common node. As will become clear in the next section, in which we consider the assembly of element equations, our goal is to generate a global system of equations applicable to the entire structure. As far as the entire structure is concerned, node 9 has an applied load of 1000 lb in the -y direction. Thus, it is immaterial how we assign nodal loads to the elements as long as the total load at the node is equal to the applied load. Keeping this in mind, when computing element equations, we can simply ignore concentrated loads applied at the nodes and apply them directly to the global equations at the start of the assembly process. Details of this process are presented in a following section. ~Assuming nod'!JJQ.ads are tQ.~dedgU:~£:Jly~t.Q..!h£.g12Qe1.~q1!,gJjQ!1~~,!h~.1injj:e el~ent equ~!~ons!2E. c::!.~ment 14 ar£§:§...f9JIRWJ/;,.
MathematicafMATLAB Implementation Plane truss element equations 1.1.2
_
-
[0] 0
0 ' 0
:n..l on the Book Web Site:
Triangular Element for Two-Dimensional Heat Flow
Consider the problem of finding steady-state temperature distribution in long chimneylike structures. Assuming no temperature gradient in the longitudinal direction, we can talce a unit slice of such a structure and model it as a two-dimensional problem to determine the y). Using conservation of energy on a differential volume, the following temperature governing differential equation can easily be established.:
T(x,
_. axa (aT) kx ax + aya (ky aT) ay + Q = 0 where kx and kyare thermal conductivities in the x and y directions and Q(x, y) is specified heat generation per unit volume. Typical units for k are W/m- °C or Btu/hr· ft· OF and those for Q are W1m3 or Btu/hr . ft3. The possible boundaryconditions are as follows: (i) Known temperature along a boundary:
T
= To specified
(ii) Specified heat flux along a boundary:
7
8
FINITEELEMENTMETHOD: THE BIG PICTURE
y (m) 0.03
0.015
qo
o
To x (m)
o
0.06
0.03 n
n
Figure 1.5. Heat flow through an L-shaped solid: solution domain and unit normals
where nx and ny are the x and Y, components of the outer unit normal vector to the example): boundary (see Figure 1.5 for
an
Inl = ~
n; + n; = 1
On an insulated boundary or across a line of symmetry there is no heat flow and thus qo = O. The sign convention for heat flow is that heat flowing into a body is positive and that flowing out of the body is negative. (iii) Heat loss due to convection along a boundary:
st
(aT
aT) =h(T -
-k an == - kx ax nx + ky ay ny
Too)
where h is the convection coefficient, T is the unknown temperature at the boundary, and Too is the known temperature of the surrounding fluid. Typical units for h are W/m2 · ·C and Btulhr· ft2 • "P, As a specific example, consider two-dimensional heat flow over an L-shaped body shown in Figure 1.5. The thermal conductivity in both directions is the same, kx = ky =
DISCRETIZATION AND ELEMENTEQUATIONS
45 Wlm . °C. The bottom is maintained at a temperature of To = 110°C. Convection heat loss takes place on the top where the ambient air temperature is 20°C and the convection heat transfer coefficient is h = 55 W/m 2 • ·C. The right side is insulated. The left side is subjected to heat flux at a uniform rate of qo = 8000 W/m2 . Heat is generated in the body at a rate of Q = 5 X 106 W1m3 . Substituting the given data into the governing differential equation and the boundary conditions, we see that the temperature distribution over this body must satisfy the following conditions: 2
2
(aax + aay
Over the entire L-shaped region
45
On the left side (lix = -1, ny = 0)
_ (45 aT (-1»)
On the bottom of the region
T
On the right side (nx
= 1, ny = 0)
On the horizontal portions of the top side (nx = 0, ny = 1) On the vertical portion of the top side (nx = 1, l1y = 0)
;
ax
= 110
; )
+ 5 X 106 =0
= 8000 =>
along y
aT
ax
= 8000 45
along x
=0
=0
ax = 0 along x = 0.06 aT ) ei 55 - ( 45 ay (1) = 55(T - 20) => ay =- 45 (T aT
- ( 45
aT
ax (1)) = 55(T -
20) =>
et
55 ax = - 45 (T -
20) 20)
Clearly there is little hope of finding a simple function T(x, y) that satisfies all these requirements. We must resort to various numerical techniques. In the finite element method, the domain is discretized into a collection of elements, each one of them being of a simple geometry, such as a triangle, a rectangle, or a quadrilateral. A triangular element for solution of steady-state heat flow over two-dimensional bodies is shown in Figure 1.6. The element can be used for finding temperature distribution
y
------x Figure 1.6. Triangular element for heat flow
9
10
FINITE ELEMENTMETHOD: THE BIG PICTURE
over any two-dimensional body subjected to conduction and convection. The element is defined by three nodes with nodal coordinates indicated by (xI' YI)' (Xz' Yz), and (x3' Y3)' The starting node of the triangle is arbitrary, but we must move counterclockwise around the triangle to define the other two nodes. The nodal degrees of freedom are the unknown temperatures at each node Tp Tz' and 13. For the truss model considered in the previous section, the structure was discrete to start with, and thus there was only one possibility for a finite element model. This is not the case for the two-dimensional regions. There are many possibilities in which a two-dimensional domain can be discretized using triangular elements. One must decide on the number of elements and their arrangement. In general, the accuracy of the solution improves as the number of elements is increased. The computational effort, however, increases rapidly as well. Concentrating more elements in regions where rapid changes in solution are expected produces finite element discretizations that give excellent results with reasonable computational effort. Some general remarks on constructing good finite element meshes are presented in a following section. For the L-shaped solid a very coarse finite element discretization is as shown in Figure 1.7 for illustration. To get results that are meaningful from an actual design point of view, a much finer mesh, one with perhaps 100 to 200 elements, would be required. The finite element equations for a triangular element for two-dimensional steady-state heat flow are derived in Chapter 5. The equations are based on the assumption of linear
y
Element numbers
0.03 0.025 0.02 0.015 0.01 0.005 0 0.01 .0.02
0
y
0.03
0.04
0.05
0.06
x
Node numbers
0.03 0.025 0.02 0.015 0.01 0.005
21 20
0
1 0
6 0.01
0.02
16
11
0.03
0.04
0.05
19 0.06
x
Figure 1.7. Triangular element mesh for heat flow through an L-shaped solid
DISCRETIZATION AND ELEMENT EQUATIONS
temperature distribution over the element. In terms of nodal temperatures, the temperature distribution over a typical element is written as follows:
where
The quantities Ni , i = 1, 2, 3, are known as interpolation or shape functions. The.superscript T over N indicates matrix transpose. The vector d is the vector of nodal unknowns. The terms b l , c I ' ... depend on element coordinates and are defined as follows:
b3
=xI
CI=X3-X2;
C2
II =XiY3 - x3Y 2 ;
12 =X3YI
-X3;
-XIY3;
C
=YI
-
Y2
= X2 - X j 3
13 = X IY2
- X2YI
The area of the triangle A can be computed from the following equation:
where det indicates determinant of the matrix. A note on the notation employed for vectors and matrices in this book is in order here. As an easy-to-remember convention, all vectors are considered column vectors and are denoted by boldface italic characters. When an expression needs a row vector, a superscript T is used to indicate that it is the transpose of a column vector. Matrices are also denoted by boldface italic characters. The numbers of rows and columns in a matrix should be carefully noted in the initial definition. Remember that, for matrix multiplication to make sense, the number of columns in the first matrix should be equal to the number of rows in the second matrix. Since large column vectors occupy lot of space on a page, occasionally vector elements may be displayed in arow to save space. However, for matrix operations, they are still treated as column vectors. As shown in Chapter 5, the finite element equations for this element are as follows:
11
12
FINITE ELEMENTMETHOD:THE BIG PICTURE
where kx = heat conduction coefficient in the x direction, ky = heat conduction coefficient in the y direction, and Q = heat generated per unit volume over the element. The matrix Je" and the vector take into account any specified heat loss due to convection along one or more sides of the element. If the convection heat loss is specified along side 1 of the element, then we have
r"
Convection along side 1:
2 1 0) [
k = hL 12 1 2 0 . "6 ' 000
hTooL12 [ .11) r" -- --2-
o
where h = convection heat flow coefficient, Too = temperature of the air or other fluid surrounding the body, and L 12 = length of side 1 of the element. For convection heat flow along sides 2 or 3, the matrices are as follows:
e" =hL23[~ ' 6
J.
Convection along side 2:
Convection along side 3:
I'
0 2 0 1 0 0 1 0
,e" ~hI,,[~ 6
n ~);
r" = hT~~3
0)
- hT-ooL31 r,,? - [~) -
1
where L23 and L31 are lengths of sides 2 and 3 of the element. The vector rq is due to possible heat flux q applied along one or more sides of the element:
Applied flux along side 1:
r, ~ q~" UJ
Applied flux along side 2:
r,~ qi'[!j
Applied flux along side 3:
r,~ qi'
m
If convection or heat flux is specified on more than one side of an element, appropriate matrices are written for each side and then added together. For an insulated boundary q = 0, and hence insulated boundaries do not contribute anything to the element equations.
DISCRETIZATION AND ELEMENT EQUATIONS
As mentioned in the previous section, we cannot solve for nodal temperatures by simply solving the equations for one eiement. We must consider contributions of all elements and specified boundary conditions before solving for the nodal unknowns. These procedures are discussed in detail in later sections in this chapter.
Example 1.2 Write finite element equations for element number 20 in the finite element model of the heat flow through the L-shaped solid shown in Figure 1.7. The element is situated between nodes 4, 10, and 5. We can choose any of the three nodes as the first node of the element and define the other two by moving counterclockwise around the element. Choosing node 4 as the first node establishes line 4-10 as the first side of the element, line 10-5 as the second side, and line 5-4 as the third side. The origin of the global x-y coordinate system can be placed at any convenient location. Choosing node 1 as the origin, the coordinates of the element end nodes are as follows: Node 1 (global node 4) = (O., 0.0225) m; Node 2 (global node 10) = (O.015, 0.03) m;
XI
=0.;
YI
x2
= 0.015; = 0.;
Y3
x3
Node 3 (global node 5) = (O., 0.03) m;
=0.0225
Y2 = 0.03
= 0.03
Using these coordinates, the constants bi' ci , and I, and the element area can easily be computed as follows: b, =0.;
c I = -0.015; II
b 3 = -0.0075
= 0.0075; c2 = 0.; 12 =0.; b2
= 0.00045;
c3 =0.015
13 = -0.0003375
Element Area.= 0.00005625 From the given data the thermal conductivities and heat generated over the solid are as follows:
Q = 5000000 Substituting these numerical values into the element equation expressions, the matrices lck and r Q can easily be written as follows: 45.
lck =
(
O. -45.
93.75]
"a = ( 93.75 93.75
There is an applied heat flux on side 3 (line 5-4) of the element. The length of this side of the element is 0.0075 m and With q = 8000 (a positive value since heat is flowing into the body) the r q vector for the element is as follows: Heat flux on side 3 with coordinates ({O., 0.0225) L
= 0.0075; q = 8000
(O., 0.03)),
13
14
FINITE ELEMENTMETHOD:THE BIG PICTURE
rq
30.) 30.
=[ a ,
The side 2 of the element is subjected to heat loss by convection. The convection term and a vector Substituting the numerical values into the formulas, generates a matrix these contributions are as follows:
kh
rho
Convection on side 2 with coordinates ((0.015, 0.03) L = 0.015; h = 55; Too = 20
(0.,0.03}),
kh=[~a 0.1375 ~.275 0.275 ~.1375);rh=[8.~5)' 8.25 Adding matrices kk and k h and vectors r Q , rq , and rh , the complete element equations are as follows:
45. O. O. 11.525 [ -45. -11.1125
-45. -11.1125 56.525
)[TT
4 )
IO
Ts
=
[123.75) 102. 132.
• MathematicalMATLAB Implementation 1.2 on the Book Web Site: Triangular element for heat flow 1.1.3 General Remarks on Finite Element Discretization The accuracy of a finite element analysis depends on the number of elements used in the model and the arrangement of elements. In general, the accuracy of the solution improves as the number of elements is' increased. The computational effort, however, increases rapidly as well. Concentrating more elements in regions where rapid changes in solution are expected produces finite element discretizations that give excellent results with reasonable computational effort. Some general remarks on constructing good finite element meshes follow. 1. Physical Geometry of the Domain. Enough elements must be used to model the physical domain as accurately as possible. For example, when a curved domain is to be discretized by using elements with straight edges, one must use a reasonably large number of elements; otherwise there will be a large discrepancy in the actual geometry and the discretized geometry used in the model. Figure 1.8 illustrates error in the approximation of a curved boundary for a two-dimensional domain discretized using triangular elements. Using more elements along the boundary will obviously reduce this discrepancy. If available, a better option is to use elements that allow curved sides. 2. Desired Accuracy. Generally, using more elements produces more accurate results. 3. Element Formulation. Some element formulations produce more accurate results than others, and thus formulation employed in a particular element influences the number of elements needed in the model for a desired accuracy.
DISCRETIZATION AND ELEMENTEQUATIONS
Actual boundary
Figure 1.8. Discrepancy in the actual physical boundary and the triangular element model geometry
x Valid mesh
Invalid mesh
Figure 1.9. Valid and invalid mesh for four-node elements
4. Special Solution Characteristics. Regions over which the solution changes rapidly generally require a large number of elements to accurately capture high solution gradients. A good modeling practice is to start with a relatively coarse mesh to get an idea of the solution and then proceed with more refined models. The results from the coarse model are used to guide the mesh refinement process. 5. Available Computational Resources. Models with more elements require more computational resources in terms of memory, disk space, and computer processor. 6. Element Interfaces. J;:~ements are joined together at nodes (typically shown as dark circles on the finite element meshes). The solutions at these nodes are the primary variables in the finite element procedure. For reasons that will become clear after studying the next few chapters, it is important to create meshes in which the adjacent elements are always connected from comer to comer. Figure 1.9 shows an example of a valid and an invalid mesh when empioying four-node quadrilateral elements. The reason why the three-element mesh on the right is invalid is because node 4 that forms a comer of elements 2 and 3 is not attached to one of the four comers of element 1. 7. Symmetry. For many practical problems, solution domains and boundary conditions are symmetric, and hence one can expect symmetry in the solution as well. It is important to recognize such symmetry and to model only the symmetric portion of the solution domain that gives information for the entire model. One common situation is illustrated in the modeling of a notched-beam problem in the following section. Besides the obvious advantage of reducing the model size, by taking advantage of symmetry, one is guaranteed to obtain a symmetric solution for the problem. Due to the numerical nature of the
15
16
FINITE ELEMENT METHOD: THE BIG PICTURE
Figure 1.10. Unsymmetrical finite element mesh for a symmetric notched beam 501b/in2
Figure 1.11. Notched beam
finite element method and the unique characteristics of elements employed, modeling the entire symmetric region may in fact produce results that are not symmetric. As a simple illustration, consider the triangular element mesh shown in Figure 1.10 that models the entire notched beam of Figure 1.11. The actual solution should be symmetric with respect to the centerline of the beam. However, the computed finite element solution will not be entirely symmetric because the arrangement of the triangular elements in the model is not symmetric with respect to the midplane. A general rule of thumb to follow in a finite element analysis is to start with a fairly coarse mesh. The number and arrangement of elements should be just enough to get a good approximation of the geometry, loading, and other physical characteristics of the problem. From the results of this coarse model, select regions in which the solution is changing rapidly for further refinement. To see solution convergence, select one or more critical points in the model and monitor the solution at these points as the number of elements (or the total number of degrees of freedom) in the model is increased. Initially, when the meshes are relatively coarse, there should be significant change in the solution at these points from one mesh to the other. The solution should begin to stabilize after the number of elements used in the model has reached a reasonable level. 1.1.4
Triangular Element for Two-Dimensional Stress Analysis
As a final example of the element equations, consider the problem of finding stresses in the notched beam of rectangular cross section shown in Figure 1.11. The beam is 4 in thick in the direction perpendicular to the plane of paper and is made of concrete with modulus of elasticity E = 3 x 1061b/in2 and Poisson's ratio v = 0.2. Since the beam thickness is small as compared to the other dimensions, it is reasonable to consider the analysis as a plane stress situation in which the stress changes in the thickness direction are ignored. Furthermore, we recognize that the loading and the geometry are symmetric with respect to the plane passing through the midspan. Thus the displacements must be symmetric and the points on the plane passing through the midspan do not experience any displacement in the horizontal direction. Taking advantage of these simpli-
DISCRETIZATION AND ELEMENTEQUATIONS
y
Element numbers
12 10 8 6 4
2 0 0
10
20
30
40
so
0
10
20
30
40
so
x
y
12 10
8 6 4
2 0
x
Figure 1.12. Finite element model of the notched beam
fications, we need to construct a two-dimensional plane stress finite element model of only half of the beam. As an illustration, a coarse finite element model of the right half of the beam using triangular elements is shown in Figure 1.12. All nodes on the right end are fixed against displacement because of the given boundary condition. The left end of the model is on the symmetry plane, and thus nodes on the left end cannot displace in the horizontal direction. Once again, in an actual stress analysis a much finer finite element mesh will be needed to get accurate values of stresses and displacements. Even in the coarse model notice that relatively small elements are employed in the notched region where high stress gradients are expected. A typical triangular element for the solution of the two-dimensional stress analysis problem is shown in Figure 1.13. The element is defined by three nodes with nodal coordinates indicated by (XI' YI)' (x 2' Y2)' and (x3' Y3)' The starting node of the triangleis arbitrary, but we must move counterclockwise around the triangle to define the other two nodes. The nodal degrees of freedom are the displacements in the X and Y directions, indicated by u and v. On one or more sides of the element, uniformly distributed load in the normal direction qn and that in the tangential direction qr can be specified. The element is based on the assumption of linear displacements over the element. In terms of nodal degrees of freedom, the displacements over an element can be written as follows: u(x, y) = N I u l vex, y)
+ N2 u2 + N3 u3
= N I VI + N2v2 + N3v3 ul
( u(x, y) ) vex, y)
= (NI 0
0
NI
N2 0 N3 0 N2 0
~J
VI
u2 =NTd v2 u3 v3
17
18
FINITE ELEMENTMETHOD:THE BIG PICTURE
y
-------------x Figure 1.13. Plane stress triangular element
where the Ni , i = 1, 2, 3, are the same linear triangle interpolation functions as those used for the heat flow element:
C j = X3-XZ;
II
=
XZY3 - X3Yz;
C =X -X I 3;
z I z = X3Yl
C3
= Xz -Xl
I 3 =XlYz
- X lY3;
- XZYI
The element area A can be computed as follows:
Using these assumed displacements, the element strains can be written as follows:
o
0
o
b3
Cz
Cz
bz
0 c3
bz
where Ex and <;, are the normal strains in the X and Y directions and 'Yxy is the shear strain. The corresponding stresses are denoted by 0;" OJ, and T.t}" respectively. The vector of stresses is denoted by CT and is related to the strain vector through Hooke's law as follows:
DISCRETIZATION AND ELEMENT EQl'iATIONS
where C is the appropriate constitutive matrix. For homogeneous, isotropic, and elastic materials under plane stress conditions, C-
1 v v 1
E
-1- V2( 0 0
,~, ]
where E = Young's modulus and v = Poisson's ratio. The finite element 'equations for this element are derived in Chapter 7 and are as follows:
kd=rq where k is the element stiffness matrix given by
where h = element thickness. The vector rq represents equivalent nodal load due to any applied distributed loads along one or more sides of an element. For a uniformly distributed load on side 1 of the element with components q" and q/ in the normal and tangential directions of the surface, the equivalent load vector is as follows: T
rq
hL I2 =T ( l 1 xq
ll -
l1yq/
l1yq"
+ nxq/
I1xq"
- nyq/
l1yq"
+ I1xq/ 0 0)
where L I2 = ~ (x2 - xJ + (Y2 - YI)2 is the length of side 1 and I1x and l1y are the components of the unit normal to side. Note that r q is a 6 x 1 column vector. It is written as a row to save space. The components of the unit normal to the side can be computed as follows:
n x
=-Y2- Y l. ' L 12
11 Y
=__x2_-x_1 L I2
A pressure component is considered positive if it is along the positive direction of the normal or tangent to the side. As shown in Figure 1.13, while moving counterclockwise around the element, the positive normal vector points in the outward direction. The positive tangent vector is 90 counterclockwise from the positive normal vector. For a uniformly distributed load on side 2, 0
hL rTq = ~(O 2 where L 23
= ~ (x3 -
0
nX" q -
11 q Y t
x 2 )2 + (Y3 - Y2)2 is the length of side 2 and
n
= Y3 -Y2.
x
L 23
'
.;1:"3
11
Y
-x
=, - -L - -2 23.
For a uniformly distributed load on side 3, T
rq
hL31 = -2-( nxq" -
l1yq/
l1yq"
+ I1xqt 0 0
I1xq"
-
l1yqt
l1yq"
+ I1xq/ )
19
20
FINITE ELEMENTMETHOD:THE BIG PICTURE
= YI -Y3.
n x
~l'
If loads are specified on more than one side of an element, appropriate vectors are.written for each side and then added together. As mentioned with the plane truss element, any concentrated applied load at a node is added directly to the global equations during assembly. This will be illustrated in a later section. Furthermore, we cannot solve for nodal displacements by simply solving the equations for one element. We must consider contributions of all elements and specified boundary conditions before solving for the nodal unknowns. These procedures are discussed in detail in later sections in this chapter.
Example 1.3 Write finite element equations for element number 2 in the finite element model of the notched beam shown in Figure 1.12. The element is connected between nodes 4, 7, and 11. We can choose any of the three nodes as the first node of the element and define the other two by moving counterclockwise around the element. Choosing node 4 as the first node establishes line 4-7 as the first side of the element, line 7-11 as the second side, and line 11-4 as the third side. The origin of the global x-y coordinate system can be placed at any convenient location. Choosing the origin as shown in Figure 1.12, the coordinates of the element end nodes are as follows: Node 1 (global node 4) = (0., 12.) in;
Xl
Node 2 (global node 7) = (5., 9.66667) in; .
x2
=0.;
=5.; x 3 =6.;
Node 3 (global node 11) = (6., 12.) in; ,I
YI
= 12.
Y2 = 9.66667 Y3 = 12.
Using these coordinates, the constants in the B matrix can easily be computed as follows: bl
= -2.33333;
b2 =0.;
b3 = 2.33333
ci
= 1.;
c2 = -6.;
Substituting the given data, we have
A
= 7.;
h
= 4;
- 0.166667 BT
=[
E = 3000000;
V=
o
O. 0.0714286 0 0.0714286 -0.166667 -0.428571 0
Plane stress C
=
[
0.2
o
-0.428571 O.
3.125 X 106 625000. 3.125 x 106 62500~.
o
1
0.166667 0 0 0.357143 0.357143 0.16{i667
21
ASSEMBLY OF ELEMENT EQUATIONS
Thus the element stiffness matrix. is 2.60913 -0.625 -1.07143 1.25 -1.5377 ~0.625
There is an applied load in the negative outer normal direction on side 3 (nodes (11, 4}) of the element. The equivalent nodal load vector rq for the element is computed as follows: Specified load components: qn = -50; qt = 0 End nodal coordinates: «(6., l2.) (0., l2.}), giving side length L = 6 Components of unit normal to the side: Hx = 0.; Hy = 1. Using these values, we get r~
= (0.
-600.
0
0
O.
-600.)
Thus the complete element equations are as follows:
106
2.60913 -0.625 -1.07143 1.25 -1.5377 -0.625
-0.625 1.41865 2.5 -2.67857 -1.875 1.25992
-1.07143 2.5 6.42857 0 -5.35714 -2.5
1.25 -2.67857 0 16.0714 -1.25 -13.3929
-1.5377 -1.875 -5.35714 -1.25 6.89484 3.125
-0.625 1.25992 -2.5 -13.3929 3.125 12.1329
u4 v4
u7 v7 u ll v ll
• Mathematica/MATLARImplementation :R..3 on the Book Web Site: Triangular element for plane stress
1.2 ASSEMBLY OF ELEMENT EQUATIONS The finite element discretization divides a solution domain or structure into simple elements. For each element the finite element equations can be written by substituting numerical values into the formulas for appropriate element type. In the assembly process, we must put the split-up solution domain back together before proceeding with the solution. The key concept in the assembly process is that at a node common between several elementsthe nodal solution is the same for all elements sharing this node. Thus contributions to that degree of freedom from all adjacent elements must be added together. To illustrate the assembly process, consider the lO-node and 8-element finite element mesh for the heat flow problem shown in Figure 1.14. The nodal degrees of freedom are the temperatures at the nodes. Since there are 10 nodes, the global system of equations is 10 x 10. Thus we start the assembly process by initializing a 10 x 10 system of equations
O. -600. O. O. O. -600.
22
FINITEELEMENTMETHOD:THE BIG PICTURE
[
111 201 201 222 301)[T!) 232 Tz 301 232 333 Ts
= (11) 12 13
Now we consider the assembly of element 1 into the global system. This element contributes to the nodal degrees of freedom (1,2,5). Since the element involves degrees of freedom Tl' Tz' and Ts, these equations will be added to global equations 1, 2, and 5, respectively. Written in expanded form, the first equation of this element is
lIlT! + 20lTz + 30lTs = 11 . Expanding it to include al1lO degrees offreedom in the model, we have
ASSEMBLY OF ELEMENT EQUATIONS
In the global system this is the first equation, and therefore the equation can be inserted into the global system as follows: 111 201 0 0 0 0 0 0 0 .0 0 0 0 0 0 0 0 0 0 O'
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
301 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
t; T2 T3 T4
T5
T6 T? Ts T9 TIO
11 0 0 0 0 0 0 0 0 0
The other two equations for the element are expanded in a similar manner. Element Equation Number
Placing these equations in the second and fifth rows, the global equations after assembly of element 1 are as follows: 222 0 0 0 232_0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0
The above procedure of reordering and expanding element equations is quite tedious. Fortunately, it is not necessary to formally carry out these steps in detail. The appropriate locations of the entries in the global equations can be determined simply by taking the list of degrees of freedom to which the element is contributing. This list is called the location vector. For element 1 the location vector is as follows: .
23
24
FINITEELEMENTMETHOD: THE BIG PICTURE
For assembling into a global vector (right-hand side), the entries in the location vector directly indicate the locations where the corresponding element quantity will contribute:
Locations for element 1 contributions to a global vector:
[;1]
To determine the global locations of the entries in an element matrix (left-hand-side coefficient matrix), we simply take all combinations of the indices in the location vector. For the first row, the locations have the row index of 1 and column indices are 1,2, and 5. For the second row, the row index is 2 and the column indices are 1,2, and 5. For the third row, the row index is 5 and the column indices are 1,2, and 5. Thus the locations in the global matrix where the corresponding element quantities are added are as follows:
Locations for element 1 contributions to the global matrix:
[
5])
[1, 1] [1,2] [2, 1] [2,2] [5, 1]
[1, [2,5] [5,5]
[5,2]
This indicates that 111 from the element matrix goes to the location [1, 1] in the global matrix, 201 into the [1,2], etc. Clearly, this gives us exactly the same global matrix after assembly of this element as before. Each element in a finite element model is processed in exactly the same manner. As a further illustration, consider assembly of element 2. Assume that the equations for element 2 are as follows: 77 80 ;' 80 88
90 100 [ 90 100 99
)[TT
2
6
Ts
)
= [21) 22 23
The location vector and the locations to which this element contributes in the global matrix are as follows:
Element 2 location vector:
[ 2~ )
Locations for element 2 contributions to a global vector:
Locations for element 2 contributions to a global matrix:
[~)
[[2,[6,2]2] [5,2]
/
[2,6] [6,6] [5,6]
[2,5])
[6,5] [5,5]
This indicates that 77 from the element matrix is added to the location [2, 2] in the global matrix, 80 into the [2,6], etc. Thus the global equations after assembly of this element are
ASSEMBLY OF ELEMENT EQUATIONS
as follows. 111 201 0 0 301 0 0 0 0 0
201 222 + 77 0 0 232+ 90 80 0 0 0 0
0 0 0 0 0 0 0 0 0 0
301 232+ 90 0 0 333 + 99 100 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 80 0 0 100 88 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0
t, Tz · T3 T4 Ts T6 T7 Ts T9
TIO
11 12+ 21 0 0 13 +23 22 0 0 0 0
The equations for the remaining six elements can be assembled in exactly the same manner. The following examples,involving plane truss, heat flow, and plane stress elements, further ~ illustrate the assembly procedure. ~
MathematicalMATLAB Implementation 1.4 on the Book Web Site:
Finite element assembly procedure Example 1.4 Five-Bar Truss Write element equations and assemble them to form global equations for the five-bar plane truss shown in Figure 1.15. The area of cross section for elements 1 and 2 is 40 em", for elements 3 and 4 is 30 crrr', and for element 5 is 20 cnr'. The first four elements are made of a material with E = 200 GPa and the last one with E =70 GPa. The applied load P = 150 kN. Each node in the model has two- displacement degrees of freedom. They are identified by the letters u and v with a subscript indicating the corresponding node number and are shown in Figure 1.16, Without considering the specified zero displacements at the supports, the model has a total of eight degrees of freedom. Thus the global equations will be a system of eight equations in eight unknowns. \
5
l·
30----4-----::;;0
4
..
\-;;,
3 p
2
o
o
2
3
4
Figure 1.15. Five-bar plane truss
25
26
FINITE ELEMENTMETHOD: THE BIG PICTURE
Figure 1.16. Five-bar plane truss finite element model
The next step is to get finite element equations for each element in the model. We simply need to substitute the appropriate numerical values into the plane truss element equations. The concentrated nodal load is added directly into the global equations at the start of assembly. Since the load is acting downward and the displacements are assumed positive along the positive coordinates, the load at node 2 is (0, -ISO kN). Since the displacements are usually small, it is convenient to use newton-millimeters. The displacements will come ~ITIillimeters and the stresses in megapascals. . For each element we substitute the numericar-aata into the plane truss stiffness matrix and assemble them into the global equations using the assembly procedure discussed earlier. The complete computations are as follows. All numerical values are in newtons and millimeters. The specified nodal loads are as follows: Node
dof
Value
2
u2
0 -150000
112
The global equations at the start of the element assembly process are t
,I
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
uj vj
u2 112
:=:
u3 113
u4 114
The equations for element 1 are as follows: E
:=:
200000; A
:=:
4000
Element Node
Global Node Number
x
Y
1 2
1 2
0 1500.
0 3500.
xj
L
:=: 0; :=:
Yj :=: 0; x 2
~ (x2 -
X j)2
:=:
1500.; Y2
+ (Y2
:=:
3500.
- Yj)2 :=: 3807.89
0 0 0 -150000 0 0 0 0
ASSEMBLY OF ELEMENT EQUATIONS
Direction cosines'. Is
= X z L- xI
,;, 0393919' m . 's
= Yz L- YI = 0.919145
Substituting into the truss element equations, we get
[U
32600.2 76067.2 -32600.2 -76067.2] j '] [0.] 76067.2 177490. -76067.2 -177490. vI _ O. -32600.2 -76067.2 32600.2 76067.2 Uz O. [ -76067.2 -177490. 76067.2 177490. v2 O. The element contributes to [1,2, 3, 4) global degrees of freedom: Locations for element contributions to a global vector: [
~]
[1, 1] [1,2] [1,3] [1, 4]] . [2, 1] [2,2] [2, 3] [2,4] 1 ba1 matnx: [3, 1] [3,2] [3,3] [3,4] and to ago [ [4,1] [4,2] [4,3] [4,4] Adding element equations into appropriate locations, we have -76067.2 o 0 0 0 76067.2 -32600.2 32600.2 177490. -76067.2 -177490. o 0 0 0 76067.2 76067.2 0 0 0 0 -76067.2 32600.2 -32600.2 76067.2 177490. o 0 0 0 -76067.2 -177490. o o 0000 o o o o o 0 0 0 o o _0 o o 0 0 0 o o o o 0 0 0 o o o
o o o
uj VI
u2 v2 u3 v3 u4 v4
-150000.
o o o
o
The equations for element 2 are as follows: E
= 200000; A = 4000
Element Node
Global Node Number
x
1 2
2 4
1500. 5000
Y 3500. 5000
= 1500.;YI= 3500.; x2 = 5000; Yi = 5000 L = ~ (x2 - xji + (Y2 - y1i = 3807.89 Direction cosines: Is = X2 ~ X = 0.919145; ms = Y2 ~ YI. = 0.393919
xj
j
Substituting into the truss element equations, we get 177490. 76067.2 -177490. -32600.2 [U2] v2 76067.2 32600.2 -76067.2 -76067.2] 76067.2 u4 177490. -177490. -76067.2 [ 32600.2 v4 -76067.2 -32600.2 76067.2
_
-
[0.] O. O. O.
27
28
FINITEELEMENTMETHOD:THE BIG PICTURE
The element contributes to (3, 4, 7, 8} degrees of freedom: Locations '0' element contributions to a global vector: [3,4] [4,4] [7,4] [8,4]
[[3.3]
. [4,3] and to a global matnx: [7, 3] [8,3]
[3,7] [4,7] [7,7] [8,7]
[!]
[3.8 1]
[4,8] [7,8] [8,8]
Adding element equations into appropriate locations, we have 32600.2 76067.2 -32600.2 -76067.2 0 0 0 0
The remaining elements can be processed in exactly the same manner. After assembling all elements, the global equations for the model are as follows: 32600.2 76067.2 -32600.2 -76067.2 0 0 0 0
• MathematicalMATLAB Implementation 1.5 on the Book Web Site: Five-bar plane truss assembly Example 1.5 Heat Flow through a Square Duct The cross section of a 20
X 20-cm duct made of concrete walls 20 ern thick is shown in Figure 1.17. The inside surface of the duct is maintained at a temperature of 300'C due to hot gases flowing from a furnace. On the outside the duct is exposed to air with an ambient temperature of 20'C. The heat conduction coefficient of concrete is 1.4 W/m . 'C. The average convection heat transfer coefficient on the outside of the duct is 27 W1m . 'C. Because of symmetry, we can model only one-eighth of the duct as shown in Figure 1.18. There cannot be any heat flow across a line of symmetry, and hence a symmetry line is equivalent to a fully insulated boundary. The solution domain is discretized into four triangular elements. This model is very coarse and therefore we do not expect very accurate results. However, showing all calculations for a larger model will be too tedious.
ASSEMBLY OF ELEMENTEQUATIONS
Figure 1.17. Cross section of a square duct
y 0.3
3
0.25 0.2 0.15 0.1 0.05
o o0.050. 10.150.l Figure 1.18. Model of eighth of a square duct
For each element we substitute the numerical data into the triangular element equations for heat flow and assemble them.into the global equations using the assembly procedure discussed earlier. There is no heat generation, thus the Q term is O. We need to compute the lck matrix for each element. Convection boundary conditions are specified on the outside surface. Thus the convection terms in the element equations will only affect element 2. Assuming this element is defined by starting with node 2, the convection term is on side 1 of the element. On the topand bottom sides, due to symmetry, there is no heat flow. They behave as insulated sides and do not require any special consideration during the finite element solution. The complete computations of element equations and their assembly follow. The global equations at the start of the assembly of the element equations are
0 0 0 0 0 000 [ 000 000
o
0 0 0 0 0
The equations for element 1 are as follows:
lex
= 1.4; ky = 1.4; Q = 0
29
30
FINITE ELEMENTMETHOD:THE BIG PICTURE
Nodal coordinates:
Xl
Element Node
Global Node Number
x
y
1 2 3
1 2 5
O. 0.2 0.1
O. O. 0.1
= 0,;x2 = 0.2;x3 = 0.1 = 0.;Y2 = 0';Y3 = 0.1
Yj Using these values, we get
= -0.1;
b2
~
c j ::;-0.1;
c2
=-0.1;
bj
0.1;
Element area, A = 0.01 Substituting these values, we get
Complete element equations:
/
The element contributes to (1,2, 5} degrees of freedom: Locations for element contributions to a global vector:
and to a global matrix:
[1, 1] [1,2] [2, 1] [2,2] [ [5, 1] [5,2]
[~ ]
5]]
[1, [2,5] [5, 5]
Adding element equations into appropriate locations, we have
Convection on side 1 (nodes (2, 3D with h = 27 and Too = 20 End nodal coordinates: ({0.2, O.) (0.2,0.3}), giving side length L = 0.3 Using these values, we get
2.7
Te"
= [ ~.35
1.35 0]
~.7
81.]
r, = [ 8~.
~;
Complete element equations:
[ ~:~~~~~~:i~~~~ =~:~][.~]s = [~i:] -1.4
-0.7
2.1
T
The element contributes to (2, 3, 5) degrees of freedom:
Locations for element contributions to, global [2, 2] and to a global matrix: [3, 2] [ [5,2]
[2,3] [2, 5]] 3] [3, 5] [5, 3] [5, 5]
n,
v"tocm
0
31
32
FINITE ELEMENTMETHOD:THE BIG PICTURE
Adding element equations into appropriate locations, we have
• MathematicafMATLAB Implementation 1.6 on the Book Web Site: Heatflow example assembly Example 1.6 Stress Analysisof a Bracket The top surface of a thin cantilever bracket is subjected to normal pressure q = 20 lb/in", as shown in Figure 1.19. The bracket is 4 in long and is 2 in wide at the base and 1 in wide at the free end. The thickness of the bracket perpendicular to the plane of pap~r is .in. The material properties are E = 104lb/in2 and v = 0.2. A very coarse four-element finite .element model using plane stress triangular elements is shown in Figure 1.20. For each element we substitute the numerical data into the element equations given in the previous section and then assemble them into the global equations using the assembly procedure discussed earlier.
!
Figure 1.19. Cantilever bracket subjected to normal pressure
y
y
Element numbers
2 1.5
2 1.5
1
1
0.5 0
0.5 0 0
2
3
4
x
2 Node numbers 4
6
5 0
2
3
Figure 1.20. Finite element model of the cantilever bracket
4
x
ASSEMBLY OF ELEMENT EQUATIONS
The pressure will be accounted for in elements 2 and 4. Choosing node 4 as the first node for element 2, the pressure 'term will be applied to side 1 of the element. The normal pressure is in the direction opposite to the outer normal to the surface and therefore it must be assigned a negative sign. For element 4, node 6 is chosen as the first node and thus the . pressure on this element is also on side 1. The complete computations of element equations and their assembly are as follows. All numerical quantities are in pounds and inches. The equations for element 1 are as follows:
h = 0.25; E = 10000; v = 0.2 10416.7 Plane stress constitutive matrix, C =
[
2083.33 10416.7
208~.33
0 0 4166.67
o
1
Nodal coordinates: Element Node
Global Node Number
x
Y
1 2 3
1 3 4
O. 2. 2.
O. O. 1.5
xI = 0.;x2 = 2.;x3 = 2. YI = 0.;Y2 = 0';Y3 = 1.5
Using these values, we get b l=-1.5;
b2 = 1.5;
b3 = O.
c j =0.;
c2 = -2.;
Element area, A = 1.5 - 0.5 0 0.5 0 BT = 0 o. 0 -0.666667 [ 0.·· -0.5 -0.666667 0.5 Thus the element stiffness matrix is:
lc = hABCB T
:::;
976.563 0 -976.563 260.417 0 -260.417
0 390.625 520.833 -390.625 -520.833 0
O. 0 0.666667
-976.563 520.833 1671.01 -781.25 -694.444 260.417
o· 0.666667 O.
260.417 -390.625 -781.25 2126.74 520.833 -1736.11
1 -260.417 0 260.417 -1736.11 0 1736.11
0 -520.833 -694.444 520.833 694.444 0
Complete element equations: 976.563 0 -976.563 260.417 0 -260.417
0 390.625 520.833 -390.625 -520.833 0
-976.563 520.833 1671.01 -781.25 -694.444 260.417
260.417 -390.625 -781.25 2126.74 520.833 -1736.11
0 -520.833 -694.444 520-,833 694.444 0
-260.417 0 260.417 -1736.11 0 1736.11
llj
vI ll3
v3 ll4
v4
O. O. O. O. O. O.
33
34
FINITE ELEMENTMETHOD:THE BIG PICTURE
The element contributes to (1,2,5,6,7,8) degrees of freedom. 1 2 5 Locations for element contributions to a global vector: 6
[1, 1] [1,2] [2, 1] [2,2] [5, 1] [5,2] and to a global matrix: [6, 1] [6,2] [7, 1] [7,2] [8, 1] [8,2]
[1,5] [2, 5] [5,5] [6,5] [7,5] [8,5]
[1, 6] [2, 6] [5,6] [6,6] [7,6] [8,6]
7 8 [1,7] . [2,7] [5,7] [6,7] [7,7] [8,7]
[1, 8] [2, 8] [5,8] [6,8] [7,8] [8,8]
Adding element equations into appropriate locations, we have 976.563 0 0 0 -976.563 260.417
h = 0.25; E = 10000; v = 0.2 Plane stress constitutive matrix, C = .
[10416.7
2083.33 0
2083.33 10416.7 0
41J67 ]
Nodal coordinates: Element Node
Global Node Number
x
Y
"1 2 3
4 2 1
2. O. O.
1.5 2. O.
b, =2.;
b2=-1.5;
b3 = -0.5
(;1 =0.;
c2 = 2.;
c3 = -2.
= 2,;x2 = 0';X3 = O. = 1.5; Y2 = 2.; Y3 = O. Using these values, we get
Xl
Yl
0 0 0 0 0 0 0 0 0 0 0 0
ul VI U2
v2 u3 v3 U4 V4
u5 v5 U6
v6
0 0 0 0 0 0 0 0 0 0 0 0
35
ASSEMBLY OF ELEMENT EQUATIONS
Element area, A = 2. -0.3750 -0.125 0.5 0 -0.5 o J B1' = 0 O. 0 0.5 ·0 [ O. -0.125 0.5 0.5 -0.375 -0.5 Thus the element stiffness matrix is ---~---~-----~------~-----------------------~-------------------------------------------- -!
Load vector due to distributed load on side 1 (nodes (4, 2)): Specified load components: qn = -20; qt = 0 End nodal coordinates: ( {2., 1.5} (O., 2.}), giving side length L = 2.06155 Components of unit normal to the side: nx
The remaining elements can be processed in exactly the same manner. After assembling all elements, the global equations for the model are as follows: 1578.78
• MathematicafMATLAB Implementation 1.7 on the Book Web Site: Plane stress example assembly 1.3
BOUNDARY CONDITIONS AND NODAL SOLUTION ;'
A variety of boundary conditions were specified for the problems considered in the previous two sections. For the truss problem, some nodes were located at the supports, which means that the displacements at these nodes must be zero. For the heat flow problem conditions to be satisfied along the boundaries involved specified temperature, convection, insulation, or heat flux. For the stress analysis problem, on some surfaces loading was. applied while other surfaces were attached to fixed supports. In the finite element analysis, some of the boundary conditions are incorporated directly into the element equations. These are known as the natural boundary' conditions. Examples of natural boundary conditions encountered in the previous sections are convection, insulation, and heat flux boundary conditions for the heat flow problem and applied surface loading for the structural problems. A review of element equations and assembly procedures discussed in the previous sections should make it clear that these types of boundary conditions have been taken into account while formulating and assembling the element equations .. The boundary conditions that are not incorporated directly into the element equations are known as the essential boundary conditions. For heat flow problems these involve specified temperatures along some boundaries. For stress analysis problems nodal sup-
BOUNDARY CONDITIONS AND NODAL SOLUTION
purts represent essential boundary conditions. Once again, a review of element equations and assembly procedures discussed in the previous sections should clearly indicate that these types of boundary conditions have not been taken into account while formulating and assembling the element equations. . Precise reasoning for splitting the boundary conditions into two categories-natural and essential-will become clear after studying the mathematical foundations of the finite element method in the following chapter. At this stage our main concern is to see how to actually incorporate boundary conditions into the finite element equations. Obviously we must take care of all applicable boundary conditions when solving a given problem. We start with the element equations and their assembly. The boundary conditions that can be incorporated through the element equations are handled first. The remaining boundary conditions are the essential boundary conditions. that are taken into account using procedures discussed in this section. In the assembly process it is assumed that all nodal degrees of freedom (dot) are unknown, However, due to essential boundary conditions, some of these degrees of freedom may have a zero or some other specified value. Therefore, introduction of essential boundary conditions into the global equations involves substituting the known values into the vector of nodal variables and malting appropriate changes to the equations.' There are three methods for accomplishing this. The first method involves rearrangement of equations but has an advantage that the final system of equations is smaller than the global system. The second method keeps the original order of variables but does require several modifications to the equations. The third method is an approximate method but requires the fewest changes to the equations. On some computers rearranging a large system of equations is very inefficient and hence the approximate method may be more economical. For hand calculations obviously the first method is preferred. Most examples presented in this text use this first approach. The details of these methods follow.
1.3.1
Essential Boundary Conditions by Rearranging Equations
As an illustration, assume that a finite element model from a heat flow problem results in the following global systerrrof equations: .
Consider the situation when some of the nodes are located on sides over which the temperatures are known. As an illustration, say T1 = 5, T4 = -7, and Ts = O. Thus we only have three nodal unknown temperatures remaining, namely 0.,7;, and T6 • These three unknowns can be obtained by solving the corresponding equations, 2, 3, and 6. Introducing the known values in the nodal degrees of freedom vector and retaining only the equations
37
Since there are only three unknowns, these equations can be rearranged by moving the known values to the right-hand side. The first column is multiplied by 5, the fourth column by -7, and the fifth column by 0, and hence we get
Moving all constant vectors to the right-hand side, we have
[
2000 21 51 21 3000 52 51 52 6000
)[TT
2
3
)
=[110) 120 + [-50) -100 + [217) 224
T6
150 .
-250
371
Adding all terms on the right-hand-side, the final 3 X 3 system of equations is as follows:
2000 21 51 21 3000 52 [ 51 152 6000
)[TT
2
3
T6
)
= [277) 244 271
The solution of this system of equations gives (T2
-?
0.136559, T3
-?
0.0796266, T6 -? 0.0433158)
Note that in these computations, when the specified value is zero, nothing gets added to the right-hand side. Thus imposing a zero value for a nodal degree of freedom simply requires removing both the corresponding row and the column from the system of equations. In this case it is not necessary to even assemble the corresponding rows and columns and therefore such degrees of freedom can be eliminated even before the assembly process begins. This is an important observation to save computational effort, especially for structural problems where zero nodal displacements are common due to supports. You may be curious as to why we retained equations 2, 3, and 6 and removed the other three. We have three unknowns left but why can we not use any three equationsto solve for the three remaining unknowns? For a precise answer to this question one needs to study the mathematical basis of the finite element method presented in the following chapter. However, a simple reasoning based on physical grounds is as follows. When a temperature
BOUNDARY CONDITIONS AND NODAL SOLUTION
is specified at a node, there must be a corresponding unknown heat source at that node to maintain that temperature. Similarly, for a structural problem, when a displacement is specified at a node, there actually is an unknown reaction corresponding to this degree of freedom. Thus, when we insert known values into the nodal variables vector, for equations to remain consistent, unknown heat source or reaction terms must be inserted into the right-hand-side vector for every specified nodal degree of freedom. Denoting theses unknowns as R 1, R4 , and R s for the illustrative example, a consistent system of equations after introducing given essential boundary conditions is as follows:
This system represents six equations in six unknowns, and if we want, we can solve these equations directly for the six unknowns. However, since we do not need the newly introduced unknowns, it is advantageous to reduce the system of equations to three for computing the three unknown temperatures. However, the only way to solve for the remaining nodal unknowns is to use equations 2, 3, and 6. Any other combination of three equations will not yield the desired result. The first three equations, as an example, will have four unknowns R p Tz, T3 , and T6 and thus are not suitable for the required solution .
• MathematicafMATIAB Implementation 1.8 on the Book Web Site: Imposing essential boundary conditions 1.3.2 Essential Boundary Conditions by Modifying Equations The method ofrearrangement of equations to handle essential boundary conditions is attractive because the final system of equations is smaller than the global system. However, on some computers rearranging a large system of equations is very inefficient, and thus it is desirable to develop a method of incorporating essential boundary conditions that does not require rearrangement of equations. A simple alternative is to modify each equation that corresponds to a specified nodal value as follows: (i) Insert 1 at the diagonal location and zero at all off-diagonal locations in the row corresponding to the specified degree of freedom. (ii) Insert the known nodal value to the corresponding location in the right-hand side. Using this approach for the illustrative example used in the previous section, we get the following system of global equations:
39
40
FINITE ELEMENT METHOD: THE BIG PICTURE
1 0 o o 0 0 10 2000 21 31 41 51 20 21 3000 32 42 52 o 0 o 10 o o 0 001 o 50 51 52 53 54 6000
t; t; T3
T4 Ts T6
::::
5 110 120 -7 0 150
Solving this system of equations for all six nodal values gives the following solution that is exactly the same as before: {Tl
:::: 5.,
Tz :::: 0.136559, T3
:::: 0.0796266,
T4
:::: -7.,
Ts :::: 0, T6
:::: 0.0433l58}
One drawback of the procedure so far is that the resulting coefficient matrix is not symmetric anymore. For large problems this is a serious drawback. It is possible to rectify the situation by making the corresponding column entries 0 and subtracting the products of terms in these columns with the known values from the right-hand side in a manner similar to that done in the previous case:
For imposing zero values of the nodal variables the process works very well because no modifications to the right-hand side ~re necessary. However, for nonzero values the procedure loses some of its simplicity. The following procedure, though approximate, is much simpler to implement. 1.3.3
Approximate Treatment of Essent.ial Boundary Conditions
The previous two methods may. be inefficient to implement on some computers because they require several modifications to the system of equations. A simple but approximate alternative is to modify each equation corresponding to a specified nodal value as follows: (i) Insert a large number at the diagonal corresponding to the known degree of freedom. (ii) Add a value equal to this large number times the known nodal value to the righthand side. Using this approach for the illustrative example used in the previous section and with 1010 as an arbitrarily chosen large number, we get the following system of global equations:
BOUNDARY CONDITIONS AND NODAL SOLUTION
10 10 10 20 30 40 50
10 2000 21 31 41 51
20 21 3000 32 42 52
30 31 32 1010 43 53
40 41 42 43 1010 54
50 51 52 53 54 6000
T! Tz
5 X 10 10
13
120 -7xlO lO '0 150
T4 Ts T6
no
Solving this system of equations for all six nodal values gives the following solution that is essentially the same as before:
(T!
= 5., Tz = 0.136559,
T3
= 0.0796266,
T4
= -7.,
Ts
= 8.97177x10-9,
T6
= 0.0433158)
To see the reason why this procedure works, consider the first equation of the modified system: ,
'
10000000000T! + lOTz + 20T3 + 30T4 + 40Ts + 50T6 = 50000000000 The two large terms dominate the equation and contribution of remaining terms is negligible. Thus the equation is effectively
10000000000T! '" 50000000000 giving
Similar reasoning shows that the fourth equation gives T4 = -7 and the fifth equation Ts = O. The other equations give the remaining unknowns. The method gives reasonable answers as long as the large number added to the diagonal is at least an order of magnitude greater than the largest element in the coefficient matrix. Using the number 106 instead of 1010, we get the following solution that shows little more error in the specified values but may still be considered reasonable:
Computation of Reactions to Verify Overall Equilibrium
For structural problems, reactions at the supports provide a quick and simple check on the validity of the analysis. For these problems the finite element equations represent equilibrium equations. Thus for overall equilibrium the sum of applied loads must be equal and opposite to the sum of all reactions at the supports. Common blunders, such as employing wrong or inconsistent units or careless computational errors, can easily be discovered if reactions are available. The concept of equilibrium is also applicable to heat flowproblems where the total heat flowing into the body must be equal to that going out of the body. After solving for the nodal unknowns, the reactions can be computed by using the equations that were deleted when the essential boundary conditions were being imposed. Thus
41
42
FINITE ELEMENTMETHOD:THE BIG PICTURE
for the illustrative example the reactions can be computed from the first, fourth, and fifth equations. Retaining these rows and inserting all nodal values, the reactions are obtained from the following computation:
(1000 30 40
5 0.136559 0.0796266 -7 0 0.0433158
50]
10 20 30 40 31 32 4000 43 53 41 42 43 5000 54
+ 100] =(R' R4 + 130 R s + 140
Rearranging and performing computations, we get the following reactions:
R I ) [1000 R 4 = . 30 ( Rs 40
1~ ;~J
10 20 30 31 32 4000 41 42 43 5000 54
5 0.136559 0.0796266
-7
o
-
[
100) 130 140
= (4695.12) -27970.9 -229.718
0.0433158 The following examples illustrate the use of reactions in validating solution consistency.
• MathematicalMATLAB Implementation 1.9 on the Book Web Site: Imposing essential boundary conditions Example 1.7 Five-Bar Truss The following global equations for five-bar plane truss were developed in Example 1.4. Incorporate essential boundary conditions into the system and obtain the final reduced system/of equations in terms of the remaining unknowns. Solve for nodal values. Compute reactions and verify overall equilibrium. 32600.2 76067.2 -32600.2 -76067.2 0 0 0 0
Since all entries in columns (1,2,7, 8) are multiplied by zero nodal values, they can be removed. The final global system of equations is thus as follows: II {j ~.'
Computation of Reactions Equation numbers of dof with specified values: (1,2,7, 8) Extracting equations (1,2,7, 8) from the global system, we have llj
VI
[32600.2 76067.2 0 0
76067.2 297490. 0 0
-32600.2 -76067.2 -177490. -76067.2
-76067.2 -177490. -76067.2 -32600.2
0 0 -120000 0
0 -120000 0 0
0 0 297490. 76067.2
ll2
7606t] 32600.2
V2 ll3
v3 ll4
v4
[R' +0]
_ R2 +0. - R3 +0. R4 +0.
44
FINITEELEMENTMETHOD:THE BIG PICTURE
Substituting the nodal values and rearranging,
R1] R2 R [ 3 R4
_
-
[32600.2 76067.2 0
0
76067.2 297490.
o o
-32600.2 -76067.2 -177490. -76067.2
o o
-76067.2 -177490. -76067.2 -32600.2
o
o o
-120000
o o
-120000
o
297490. 76067.2
o o
7606~'2] 32600.2
0.538954 -0.953061 0.264704 -0.264704
o o
Carrying out computations, the reactions are as follows: Label
dof
Reaction
RI R2 R3 R4
ul vI
54926.7 159927. -54926.7 -9926.67
u4 v4
Sum of reactions: dof: u dof: v
0 150000.
There is no applied load in the x direction. Since the sum of reactions in the horizontal direction is zero, the equilibrium is satisfied in this direction. There is an applied load of 150000 in the -y direction which balances with the sum of reactions in the y direction, indicating that equilibrium is satisfied in this direction as well. Hence the solution satisfies the overall equilibrium. /
Example 1.8 Heat Flow through a Square Duct The following global equations for heat flow through a square duct are developed in Example 1.5. Incorporate essential boundary conditions into the system and obtain the final reduced system of equations in terms of the remaining unknowns. Solve for nodal values. Compute reactions and verify overall energy balance: 1.4 0 0 -0.7 -0.7
0 4.56667 1.58333 0 -2.1
0 1.58333 3.51667 0.35 -1.4
-0.7 0 0.35 3.15 -2.8
-07] T'] 8~] -2.1 -1.4 -2.8 7.
T2 T3 = 81. T4 0 Ts
0
Nodes 1 and 4 are on the inside face that is maintained at a temperature of 300·C. Thus the essential boundary conditions are as follows: Node
1 4
dof
Value 300 300
BOUNDARY CONDITIONS AND NODALSOLUTION
Incorporating these boundary conditions means setting TI tions 1 and 4 from the global system:
Solving the final system of global equations, we get the following nodal values: (Tz = 93.5466, T3
= 23.8437, t; = 182.833}
Complete Table of Nodal Values
T I
2 3 4 5
300 93.5466 23.8437 300 182.833
Computation of Reactions Equation numbers of dof with specified values: (1, 4) Extracting these equations from the global system (the one prior to adjustment for essential boundary conditions), we have
Sum of reactions: 313.431 The sum of reactions represents the total heat energy put into the system. This must be balanced by the heat loss due to convection. The convection heat loss takes place only on side 2-3 of the model. The average temperature of this side is
Tavg = !(T2 + T3 ) = 58.695 Assuming this temperature to be constant for the entire O.3-m length of the side, Convection heat loss from side 2-3 = hL(Tavg - T,,,)
= 313.43
This heat loss is exactly the same as the sum of reactions, indicating that the solution satisfies overall energy balance.
Example 1.9 Stress Analysis of a Bracket The following global equations for stress analysis of a bracket are developed in Example 1.6. Incorporate essential boundary conditions into the system and obtain tllli" final reduced system of equations in terms of the remaining unknowns. Solve for nodal values. Compute reactions and verify overall equilibrium: 1578.78 195.313 -276.693
Nodes 1 and 2 are on the fixed end of the bracket. Thus the essential boundary conditions are as follows: Node
dof
Value
1
uj vj u2
0 0
2
v2
0 0
47
BOUNDARY CONDITIONS AND NODALSOLUTION
Incorporating these boundary conditions means removing equations 1,2,3, and 4 from the global system. Furthermore, since the specified values are 0, the corresponding columns carralso be removed. Thus, after incorporating the essential boundary conditions, the final system of equations is as follows: 3125. -520.833 -1171.88 520.833 -651.042 260.417 -325.521 -781.25
Computation of Reactions - . Equation numbers of dof with specified values: {I, 2, 3, 4} Extracting equations {I, 2, 3, 4} from the global system, we have "I VI
U2 V2
[""'"
195.313 -276.693 325.521
195.313 1725.26 65.1042 -1204.43
-276.693 65.1042 1253.26 -585.938
325.521 -1204.43 -585.938 1595.05
-976.563 520.833 0 0
260.417 -390.625 0 0
-325.521 -781.25 -976.563 260.417
-781.25 -130.208 520.833 -390.625
0 0 0 0
0 0 0 0
0 0 0 0
!] ". ['", ] "3
_ Rz + O. - R3 -1.25 R. -5. v.
V
3
"5 Vs
"6 v6
48
FINITEELEMENTMETHOD: THE BIG PICTURE
Substituting the nodal values and rearranging, o o o o RI )
The sum of reactions must be balanced by the applied loading. The loading is applied normal to the top surface. To determine the total load and its components in the horizontal and vertical directions, we need its length and normal vector. The end nodal coordinates of this line are Yz
=2
The length and components of the unit normal to the line are computed as follows: L =
~(xz _xj)z + (yz -
Yjf= 4.12311
nx = (yz - Yj)/L = 0.242536;
ny = -(xz - xj)IL = 0.970143
The total applied load and its components in the coordinate directions can now be determined as follows:
These are equal and opposite to the sum of reactions. Hence the solution satisfies the overall equilibrium.
ELEMENTSOLUTIONS AND MODELVALIDITY
1.4
ELEMENT SOLUTIONS AND MODEL VALIDITY
Once the solution for all nodal degrees of freedom is available, we can go back to the element equations and develop a complete solution over each element. The process simply requires substituting computed nodal values into the assumed element solution defined during derivation of element equations. Any secondary quantities, such as derivatives or integrals involving the solution, can be obtained by performing necessary computations on the element solutions. The examples presented in the following sections clarify the procedure for different elements introduced earlier in this chapter.
1.4.1
Plane Truss Element
As will be seen in Chapters 2 and 4, the fundamental assumption made in deriving the plane truss element equations is that the axial displacement varies linearly over the element. This means that, in terms of nodal degrees of freedom, the displacement along the element axis is written as follows: O:5,S:5,L
where, as shown in Figure 1.21, s is a local coordinate that runs along the element length with s = 0 at the first node of the element to s = L at the second node and L = length of the element. The terms d l and dz are the axial displacements at the element ends. The relationship between the local axial displacements and the global nodal degrees offreedom is as follows.
Axial displacements:
(~~) = (~
Transformation matrix:
T-Cs - 0
Ins
0
0
Is
Ins
0
0
Is
o
m')
["'~] VI
= Td
~J
where Is and Ins are the direction cosines of the element axis as defined earlier. The axial strain is simply the first derivative of the axial displacement, giving constant strain over the element as follows:
Using Hooke's law, the axial stress is obtained by multiplying strain by the modulus of elasticity: (F
= EE
49
50
FINITE ELEMENTMETHOD:THE BIG PICTURE
y Local dof
Global dof
x Figure 1.21. Global and local degrees of freedom for a plane truss element
Since axial stress is equal to axial force divided by the area, the axial force in the element is
F
=erA
The sign convention used in these equations assumes that the tension is positive and the compression is negative. Example 1.10 Five-Bar Truss The following nodal values for a five-bar plane truss were obtained in Example 1.7. Determine axial strains, axial stresses, and axial forces in different elements of the truss. ,/
1 2 3 4
u
v
0 0.538954 0.264704 0
0 -0.953061 -().264704 0
The displacements are in inches, loads in pounds, and stresses in pounds per inch squared. The solution for element 1 is as follows: Element nodal coordinates: first node (node #1): (0,0); Second node (node #2): (1500.,3500.)
= 0; YI = 0; x2 = 1500.; Y2 =3500. L = ~ (x2 - x l )2 + (Y2 - YI)2 = 3807.89 Direction cosines: Is = x 2 - Xl = 0.393919; Ins = Y2 - YI = 0.919145 . L L
Xl
....
,
. r:
.:
ELEMENT SOLUTIONS AND MODEL VALIDITY
Global to local transformation matrix, T _ (0.393919
-
0
Element nodal displacements in
0.919145 0 0 ) 0 0.393919 0.919145
gl~al
coordinates, d
=
[~] =[ 0.53~954 v2
]
-0.953061
Ele~en~ nodal displacements in 10Calc~~~i~.~!eS, dL=':.~~-0.6~·3697 ) £. I Axial displacements at element ends, d~ ~d? = ::Q:!!63§27J':"d. e if'0 Axial strain, (d d, )/L -0.000174'295 r;(C.'-J cL,._.j I) (1- Cd q~\.l \L E'=
2 -
=
E=200000;A=~ I '/" Axial stress, ap.:= J{@!: -34.8591; Axial force = (J"A = -139436.. v
.
~
i)
'
For any other elementthe calculations follow exactly the same pattern. The solution summary is as follows:
1 2 3 4 ,5 ~
Stress
Axial force
-34.8591 -629994 -10.5881 -10.5881 22.4608
-139436. -25199.8 -31764.4 -31764.4 44921.7
MathematicafMATLAB Implementation 1.10 on the Book Web Site:
Plane truss element results ~
MathematicafMATLAB Implementation 1.11 on the Book Web Site: Complete solution ofa plane truss
1.4.2 Triangular Element for Two-Dimensional Heat Flow The finite element equations for a triangular element for two-dimensional steady-state heat flow are based on the assumption of linear temperature distribution over the element. In terms of nodal temperatures, the temperature distribution over a typical element is written as follows:
51
52
FINITE ELEMENTMETHOD:THE BIG PICTURE
where
CI = X 3 - X 2;
11 = X 2Y3 -
X 3Y2;
C2 = X t - X 3;
C3 =X2 -XI;
12 =x 3Yt -
13 =XI Y2 -
x 1Y3;
X2YI;
The quantities N i , i = 1, 2, 3, are linear triangle interpolation functions. The vector d is the vector of nodal quantities that are all known at this stage. Knowing the coordinates at' the three element nodes, the interpolation functions can easily be written for each element. Multiplying these interpolation functions by the nodal temperature computed from the solution of global equations gives the temperature distribution T(x, y) over each element. This expression can be differentiated or integrated in a usual manner to obtain other quantities of interest. Note that for each element a different expression for the solution T(x, y) is obtained. The finite element assembly process matches the degrees of freedom at the common nodes between the elements. Therefore, T must be continuous across common element boundaries. However, since T is a linear function of X and y, its derivatives are constant. Thus over each element the derivative values could be different, and in general two different derivative values will be computed at a common interface between the two elements. The magnitude of this discontinuity can be used as a guide to determine if the finite element discretization is suitable or needs further refinement. Example 1.11 neat Flow through a Square Duct The following nodal temperature values for heat flow through a square duct were obtained in Example 1.8. Determine the temperature distribution over each element and its x and y derivatives. By comparing the magnitudes of these derivatives across common element interfaces, comment on suitability of the finite element mesh. Node
Temperature
1 2 3 4 5
300 93.5466 23.8437 300 182.833
The temperature is in degrees Celsius and heat flow in watts per meter. The solution for element 1 is as follows: Element nodes: First node (node #1): {O., O.}; Second node (node #2): {O.2, O.}; Third node (node #5): {O.I, O.I} XI = 0.;x2 = 0.2;x3 = 0.1 YI
= 0.;Y2 = 0';Y3 = 0.1
ELEMENT SOLUTIONS AND MODEL VALIDITY
Using these values, we get
= -0.1; = -0.1; II = 0.02;
Element area, A
bl
bz = 0.1;
cj
C
z = -0.1;
I z =0.;
b3 = O.
c3
=0.2
u, = O.
=.0.01
Substituting these into the formulas for triangle interpolation functions, we get Interpolation functions, NT = (-5.x - 5.y + 1., 5.x - 5.y, 1O.y) From a global solution the temperatures at the element nodes are (from nodes (I, 2, 5)), d T
= {300, 93.5466, 182.833}
Thus the temperature distribution over the element, T(x, y) 139.406y + 300. Differentiating with respect to x and y, aT/ax
= NTd = -1032.27x-
= -1032.27 and oT/oy = -139.406
For any other element the calculations follow exactly the same pattern. The solution summary is as follows:
As seen from the model shown in Figure 1.18, elements 1 and 2 have a common side between them. However, the derivative values for the two elements show significant difference. Elements 3 and 4 also share a side. The x derivatives for the two elements are the same; however, their y derivatives are vastly different. TIns large difference in the derivative values shows thatthe finite element mesh used for the analysis is very coarse. A much finer mesh is needed for any meaningful results. As mentioned earlier, this coarse mesh is used here to illustrate all necessary computations in detail. ~
MathematicafMATLAB Implementation 1.12 on the Book Web Site:
Heat flow element results .
• MathematicalMATLAB Implementation 1.13 on the Book Web Site: Complete solution of a heat flow problem
53
54
FINITEELEMENTMETHOD:THE BIG PICTURE
1.4.3 Triangular Element for Two-Dimensional Stress Analysis The derivation of the plane stress triangular element equations is based on the assumption of linear displacements over the element. In terms of nodal quantities, the displacements over an element can be written as follows.
=N, u + N2u2 + N3u3 V(X,y) = Njv I +N2v2 +N3v3
U(x, y)
j
Uj
( U(X, y) ) V(X,y)
=( n, 0
0 Nj
N2
0
N3
0
N2
0
~J
vI U2'
=.NTa
V2 U3 V3
where Nj , i = 1, 2, 3, are the triangle interpolation functions (same as those used for the triangular element for heat flow). Knowing the coordinates at the three element nodes, the interpolation functions can easily be calculated for each element. Multiplying these interpolation functions by the nodal displacements computed from the solution of global equations gives the horizontal and vertical displacement distributions u(x, y) and vex, y) over each element. These expression can be differentiated or integrated in a usual manner to obtain other quantities of interest. ' . As discussed later in Chapter 7, the three components of in-plane element strains are computed directly by differentiating displacement functions as follows:
o cj bj
b2 0 c2
0 c2 b2
b3 0 c3
The stresses can be computed as follows:
a =C€ where C is the appropriate constitutive matrix. For homogeneous, isotropic elastic materials under plane stress conditions
C=
E
1-
1 v v 1
[o 0
0] 0
l;:v
where E = Young's modulus and v = Poisson's ratio. For plane stress' problems, the three out-of-plane strain and stress components are as follows:
ELEMENTSOLUTIONS AND MODEL VALIDITY
v(~
+ oy)
EZ=-·E;
'YyZ
=0;
~=O;
'Yix = 0 TZ.T
=0
The complete vectors of strains and stresses are ordered as follows:
(J"
= (~ oy
From a design point of view, one is often interested in principal stresses or effective (von Mises) stress. The principal stresses can be computed by determining eigenvalues of the following 3 x 3 matrix of stress components:
Principal stresses
= eigenvalues[S]
The effective (von Mises) stress can be computed from the component stresses using the following equation:
a;, =
~~ (a:T -
oy)2
+ (oy -
~)2 + (~- ~)2 + 6(-s, + T;z + T~T)
Similar to the heat flow problem, for each element different expressions for displacements u(x, y) and vex, y) are obtained. The finite element assembly process matches the degrees of freedom at the common nodes between the elements. Therefore, the displacements must be continuous across common element boundaries. However, since displacements are linear functions of x and y, their derivatives, and hence strains and stresses, are constant. Thus over each element the stress values could be different, and in general two different stress values will be computed at a common interface between the two elements. The magnitude of this discontinuity can be used as a guide to determine if the finite element discretization is suitable or needs furtherrefinement,
Example 1.12 Stress Analysis of a Bracket The following nodal displacement values for the bracket problem were obtained in Example 1.9. Determine displacement distribution over each element. Using these, compute element strains, stresses, principal stresses, and effective stress. By comparing the magnitudes of stresses across common element interfaces, comment on the suitability of the finite element mesh.
1 2 3 4 5 6
u
v
0 0 -0.0103553 0.00472765 -0.0131394 0.0000838902
0 0 -0.0255297 -0.0247357 -0.0554931 -0.0555664
55
56
FINITEELEMENTMETHOD:THE BIG PICTURE
Solution for Element 1 h
= 0.25; E = 10000; v = 0.2
Plane stress constitutive matrix, C
=[
10416.7 2083 0.33
2083.33 10416.7
o
0 1 0 4166.67
Element nodes: first node (node 1): {O., O.}; second node (node 3): {2.,O.}; third node (node 4): {2., 1.5}
=0.;
X2
=2.;
YI =0.;
Y2
=0.;
XI
Using these values, we get
b2
b3
c I =0.;
= 1.5; c2 = -2.;
II
12 =0.;
13 =0.
bl
Element area: A
BT =
= -1.5; 3.;
= O. c3 = 2.
= 1.5
~.
- 0.5 0 [ O. -0.5
~.5
-0.666667
-~.666667~· 0.5
~0. 6666671"\
0.666667
)
Substituting these into the formulas for triangle interpolation functions, we get /
From a global solution the displacements at the element nodes are {displacements at nodes (I, 3, 4)): dT ={O,0, -0.0103553, -0.0255297, 0.00472765, -0.0247357} The displacement distribution over the element is
Substituting these stress components into appropriate formulas, Principal stresses
= (0
-2.72856
-55.3749)
Effective stress (von Mises) = 54.0623 Proceeding in exactly the same manner, we can get solutions for all elements. A summary of the solution at element centers follows:
1
Coordinate
Displacement
Stresses
Principal Stresses
Effective Stress
1.33333 0.5
-0.00187588 -0.0167551
-52.8309 -5.27256 0 -11.2898 0
0 -2.72856 -55.3749
54.0623
0
2
0.666667 1.16667
0.00157588 -0.00824522
24.6232 4.92464 0 -51.5326 0 0
67.2393 0 -37.6915
92.0659
3
3.33333 . 0.333333
-0.0078036 -0.0455297
-14.6533 -3.66334 0 -7.32667 0 0
0 0 -18.3167
18.3167
4
2.66667 0.833333
-0.00184791 -0.0352772
3.10223 5.91407 0 -21.7822 0 0
26.3357 0 ·-17.3194
38.0742
As seen from the model shown in Figure 1.26, elements 1 and 4 have a common side between them. However, the effective stress values for the two elements show significant difference. Elements 3 and 4 also share a side. The effective stresses for these two elements are quite different as well. This large difference in the stress values shows that the finite
57
58
FINITE ELEMENTMETHOD: THE BIG PICTURE
element mesh used for the analysis is very coarse. A much finer mesh is needed for any meaningful results. As mentioned earlier, this coarse mesh is used here to illustrate all necessary computations in detail.
• MathematicafMATLAB Implementation 1.14 on the Book Web Site: Plane stress element results
• MathematicafMATLAB Implementation 1.15 on the Book Web Site: Complete solution ofa plane stress problem 1.5
SOLUTION OF LINEAR EQUATIONS
As seen from the previous section, the global system of equations after boundary conditions consists of n equations in n unknowns:
Kd=R Most scientific calculators have functions for solving linear system of equations. These are useful for solving small textbook problems. Computer algebra systems (Mathematica, MATLAB, MAPLE, etc.) have sophisticated built-in functions for solving large systems of linear equations. For example, using Mathematica, a system of equations can be solved by first defining matrix K and right-hand-side vector R and then issuing the following command: LinearSolve [K, RJ
For completeness two basic methods fbr solution of a large-scale linear system of equations are briefly discussed in this section. Detailed treatment of these methods is beyond the scope of this text.
1.5.1
Solution Using Choleski Decomposition
An efficient method to solve the system of equations is to first decompose the matrix K into a product of lower and upper triangular matrices. For symmetric matrices the socalled Choleski decomposition is very efficient. In this decomposition a symmetric matrix K is decomposed into a product of a lower triangular matrix L and its transpose L T in the following form:
K=LL T
[,n
0
121
122
0 0
L = 1~1
1~2
1~3
.;
.:
In3
0]o . 0
1~,
'
LT
=
III 0 0
121 122
131 132
0
1~3
0
0
0
'., ] In2 In3
l~n
SOLUTION OF LINEAR EQUATIONS
Such a decomposition is always possible for nonsingular symmetric matrices. Assuming this decomposition is known, wecan solve the system of equations in two steps as follows: Kd
If we define LTd
T
= R =? LLT d = R
=y, then
LL d
1
,
0
0
1:: l 0
= R =? Ly = R
zz
1~1 1~2 133
=?
[
1111 l l1z 1113 Since L is a lower triangular matrix, the first row gives the solution for the first variable as YI = r/l l l · Knowing this, we can solve for Yz = (r z - lzIYI)llzz from the second equation. Thus, proceeding forward from the first equation, we can solve for all intermediate ' variables Yi' This is known asforward elimination. _ ri
"i-II
- LJk-1 ikYk.
1..
Yi -
i = 1,2, ... , Ii
'
II
Now we can solve for the actual unknowns di as follows:
LTd =y
=?
[In_~ 0
lz1 131 lzz 132 0 1~3 :
0
0
... ... ...
1.,]["']
...
1 ~1I d
'.
dz
d3 = ll
n Yz
l l1z
1113
'
Y3
;11
In this system the last equation has only one unknown, giving dll = y/l,l1l • The second to the last equation is used-next to get dll _ l , and thus, working backward from the last equation, we can solve for all unknowns. This is known as backward substitution. i
= n, n -
1, ... , 1
The procedure is especially efficient if solutions for several different right-hand sides are needed. Since in this case the major computational task of decomposing the matrix into LLT form is not necessary for every right-hand side, one simply needs to perfortn a forward elimination and a backward substitution for each different right-hand side. ChoJeski Decomposition The Choleski decomposition algorithm can be developed by considering direct multiplication of terms in the Land L T matrices and equating them to the corresponding terms in the K matrix:
59
60
FINITEELEMENTMETHOD:THE BIG PICTURE
J(
k 11 1'21 =LL T ::=} k~1 k"l
k21 k31 ~2 k32 k~2 k33
k"l k"2 k"3
III
=
0
0
I~"]['!'
121 122 0 I~I 1~2 1~3 :
",,2
k"3
k""
I'd
1"3
1"2
121 131 122 132
...
...
I", ] 1,,2
0
1~3 ... 1"2
0
0
...
I,:"
By direct multiplication we can see that the entries in the first row of the L matrix can be established as follows:
kl l
Row 1:
.
= Iii::=} III = {k::
_
_ k21
~I - 111/ 21 ::=} 121 -
lei! = II IIi! ::=} Ii! = _
1
-II
ki!
t:
i
= 2, ... , n
11
Once entries in row 1 are known, we can proceed to row 2 and establish the entries there by direct multiplication as follows: le22 = I~I + 1~2 ::=} 122 = ~ le22 -/~1
Proceeding in a similar manner, we canwrite the following general formulas for entries in the ith row of the L matrix: I -
Notice that the procedure requires taking the square root of terms corning from the diagonal of the J( matrix and then division by this term to establish other terms in a given row. This is called the pivot element. Clearly these operations will fail if there is a negative or zero pivot. However, as long as the matrix J( is nonsingular, it is always possible to rearrange the system of equations to avoid a negative or zero pivot. The computational steps can be written in the form of the following algorithm: Choleski Decomposition Algorithm For i = 1, n For j = 1, i-I
SOLUTION OF LINEAR EQUATIONS
end Example 1.13 Find the solution of the following system of equations using Choleski decomposition:
[; l~
6 2 3 -10
~ -l~][~~]_ [~]7 -5
11 -5
d3
21
-
a,
8
We first generate the LL T decomposition of the coefficient matrix using the above algorithm:
3
K
= LL T = 1 -{IT [
2
o
0
0
1 --{IT
o ..;7
-{7
o
o
][30 -{IT 1 20 --{IT 1] = [93 0 ..;7 00
0
{io
-..;7
6
{i
3
6 3]
3 2 -10 12 2 11 -5 21 -10 -5
Using the L matrix, the solution for intermediate variables (forward elimination) is as follows: .
Using the L T matrix, the solution for actual variables (backward substitution) is as follows: 5
3" 13
3 {IT 11
3-17 43
3-{i
61
62
FINITE ELEMENTMETHOD: THE BIG PICTURE
- (;;2d -.-:!1. ---- d - 11 -y L. 4 - 3-12 --.' 4 6
Row 4:
+ (--v'7)d I 4 =.l.L 30 ====> d 3 =
Row 3:
- '7d -y I 3
Row 2:
{fidz + Od3 + (-{fi)d 4
323 4Z
= 3~ ====> dZ = ':rt
3d l + 1dZ + 2d 3 + Id 4 = ~ ====> d, = -7i~
Row 1:
1.5.2 Conjugate Gradient Method The well-known conjugate gradient method for solution of unconstrained optimization problems can be used to develop an iterative method for solution of a linear system of equations. The method has been used effectively for solution of nonlinear finite element problems that require repeated solutions of large systems of equations. Consider a symmetric system of n X n linear equations expressed in matrix form as follows:
Kd=R
['":,1'
kZJ
k JZ
kF "liZ
k'"r] ["] 10.11
dz _ rz
I~I
.. cL
Finding the solution of this system of equations is equivalent to minimizing the following quadratic function: minf(d)
= 4dTKd -
dTR
,/
This can easily be seen by using the necessary conditions for the minimum, namely, the gradient of the function must be zero:
where
Since K is a symmetric matrix, the transpose of the first term in each row is same as the second term. Thus the gradient can be expressed as follows:
we see that the gradient of the quadratic function is g(d)
= Kd-R
A condition of zero gradient implies Kd - R = 0, which clearly is the given linear system of equations. Thus solution of a linear system of equations is equivalent to finding a minimum of the quadratic function f. Basic Conjugate Gradient Method In the conjugate gradient method, the minimum of f is located by starting from an assumed solution d(O) (say with all variables equal to zero) and performing iterations as follows:
k = 0, 1,... where a(k) is known as the step length and h(k) is the search direction. At the kth iteration the search direction h(k) is determined as follows:
The scalar multiplier fJ determines the portion of the previous direction to be added to determine the new direction. According to the Fletcher-Reeves formula,
After establishing the search direction, the minimization problem reduces to finding a(k) in order to
For the minimum the derivative of this expression with respect to zero, giving the following equation for the step length:
a(k)
must be equal to
Noting the symmetry of K, the first two terms can be combined and moved to the right-hand side to give
giving
Since f3 is usually small, h(k) "" _g(k), and thus for computational efficiency the step length expression is slightly modified as follows: "
Basic Conjugate Gradient Algorithm The computational steps can be organized into the following algorithm. Choose a starting point d(O). Compute g(O) = Kd(O) - R. Set h(O) = _g(O). Compute reO) = (g(O){g(O). Choose a convergence tolerance parameter E. Set k O. 1. If -,J-k) .s
E,
stop; d(k) is the desired solution. Otherwise, continue.
2." Compute the step length:
3. Next point:
SOLUTION OF LINEAR EQUATIONS
4. Update the quantities for the next iteration: g(k+l)
= K(d(k) + o:(k)/z(k») _ R == g(k) + o:(k)Z(k)
,.(k+l)
=
fP)
=
/z(k+1)
(g(k+l){g(k+l)
,.(k+l)h·(k)
= -s"'" + (3(k)/z(k)
5. Set k = k: + 1 and go to step 1. Example 1.14 Find the solution of the following system of equations using the basic conjugate gradient method:
Next point, d(4) == d(3) +(1'(3)h(3) == {-9.48052, 7.56061, 7.69048, 7.16667} g(4) == g(3) + (1'(3)Z(3) == (7.10543 X 10- 15, -9.76996 X 10- 15,0.,2.57572 x 10- 14 ) r(4) == g(4)Tg(4) == 8.09371 X 10- 28 /3(4) == r(4)/r(3) == 3.78081 X 10- 30 h(4)
==
_g(4)
+ /3(3)h(3)
== {~7.10543
x 10- 15,9.76996 X 10- 15,8.67954 X 10- 29, -2.57572 x 1O-14 }
Solution converged after four iterations: g == {7.10543 X 10- 15, -9.76996 X 10- 15,0.,2.57572
x 10- 14 1
X 10- 28
r == gTg == 8.09371 Solution, d == {-9.48052, 7.56061, 7.69048, 7.16667}
Preconditioned Conjugate Gradient When the coefficient matrix K is illconditioned, the convergence of the conjugate gradient method could be very slow. Introducing a suitably chosen preconditioning matrix P improves convergence. To see this, we multiply both sides of the system of equations by the inverse of the preconditioning ;' matrix to get
Clearly, if P == K, we have the solution of the system of equations. Thus, intuitively, we can expect that, if P is a matrix that is reasonably close to K, then the method will converge faster. The matrix P is known as the preconditioning matrix. Several different techniques have been proposed for establishing this matrix, Two popular choices will be discussed later in the section. First we develop the computational procedure for the solution of equations when the preconditioning matrix is included. In general, the product P"! K does not produce a symmetric matrix, even if both matrices are individually symmetric. In order to get a symmetric system of equations, we introduce the following decomposition of the P matrix,
and a new vector of intermediate variables y,
SOLUTldN OF LINEAR EQUATIONS
Introducing these, the original system of equations is modified as follows:
Thus we have a system of equations as follows:
where K =E- 1!W- T and k =E-1R. The matrix K is a symmetric matrix, At this point we can use the standard conjugate gradient method to find a solution for intermediate variables y: g(k)
=Ky(k) _ R;
j3(k)
=
ii(k)
rn(k){ -(k) l.5 g . [g(k-J){g(k-l) , y(k+l)
=y(k)
= _g(k) + j3(k)h(k-l) a(k)
=
fn(k){ -(k) g (h(k){iih(k) 15
+ a(k)h(k)
Multiplying vector y(k+l) by E- T gives the solution of the original problem:
A serious drawback of the procedure so far is that it requires a decomposition of the preconditioning matrix p-l = E-TE-l, which is a very time consuming task for a general matrix. Fortunately, it is possible to rewrite the above expressions so that only p-l appears in the equations and thus there is no need to explicitly carry out the decomposition. Substituting for K and k into the g(k) expression, we have
Similarly, other expressions in the algorithm can be rewritten as follows: , j3(kl
=
[g(k) { g(k) [g(k-J){glk-l)
a(k)
=
[g(k) { E-TE-1g(k) [g(k-J){E-TE-1g(k-l)
(g(k){g(k)
=
(h(k){E-lKE-Ti'L0') y(k+1)
=y(k)
+ a(k)h(k)
=
[g(k){p-l g(k)
.
[g(k) { p-lg(k)
= --=::...........:-:;;--=--[g(k-J){ p-lg(k-J)
r:
'[g(kl]T I g(k) = .:::::--=...,,,,--..:::-(E-Th(kl { !W-Tii(k) (h (k){ Kh(k)
67
68
FINITEELEMENTMETHOD:THE BIG PICTURE
Multiplying the expression for the intermediate variables by E-T , we have
Preconditioned Conjugate Gradient (PCG) Algorithm The computational steps can be organized into the following algorithm: Initialization
Choose a starting point d(D). Set g(D) Solve for W(D): Pw(D) = g(D). Set h(D) = -W(D). Compute
r(D)
=Kd(D) -
= (g(D){ W(D).
Choose a convergence tolerance parameter
1. If ,J.k) ~
E,
R.
E.
Set k
= O.
stop; d(k) is the desired solution. Otherwise, continue.
2. Compute the step length:
3. Next point: d(k+l)
= d(k) + a(k)h(k)
I' 4. Update the quantities for the next iteration: g(k+l)
=g(k) + a(k)z(k)
Solve for W(k+l): PW(k+l) r(k+l) j3(K)
=
=g(k+l)
(g(k+l){W(k+l)
= ,J.k+I)/,J.k)
h(k+l)
= _W(k+l) + j3(k)h(k)
5. Set k = k + 1 and go to step 1. Jacobi Preconditioning One of the simplest choices of the preconditioning matrices is the so-called Jacobi preconditioning. Here the P matrix is a diagonal matrix with entries . equal to the diagonal elements of the J( matrix. Thus p-I is established as follows: -I _ 1. Pii - k ..' II
·1 =0'' P:IJ
i =1= j;
i, j
= 1,... , n
SOLUTION OF LINEAR EQUATIONS
Example 1.15 Find the solution of the following system of equations using the peG method with Jacobi preconditioning:
9
3
3 12 6· 2 [ 3 -10
9. Preconditioning matrix, P
=.
[
~
0
0
Tl~.
0 ]
2~.
Starting point, d(O! g(O) =. Kd(O)
-R
Solving Pw(O) h(O)
=.
=. (0,0,0, O) (-5, -6, -7, -8) g(O), we have w(O) = {-0.555556, -0.5, -0.636364, -0.380952}. =.
= (-7.99361 X 10- 15, 1.42109 X 10- 14,8.88178 X 10- 15, -3.73035 x 1O-14j Solving PW(4) = g(4), we have W(4) = (-8.88178 X 10- 16,1.18424 X 10- 15,8.07435 X 10- 16, -1.77636 x 10- 15 ) r(4) =g(4)T w(4) = 9.73646
X
10- 29
/3(4)
= r(4)jr<3) = 6.29497 x 10- 30
h(4)
= _g(4) + /3(3lh<3)
= (8.88178 X 10- 16, -1.18424 X 10- 15 , -8.07435 X 10- 16,1.77636 x 1O-15j The solution converged after four iterations: g
= (-7.99361 X 10- 15, 1.42109 X 10- 14,8.88178 X 10- 15, -3.73035 x 1O-14j
r
=gTg = 9.73646 X 10- 29
Solution, d = (-9.48052,7.56061,7.69048, 7. 16667j I
Incomplete Choleski Preconditioning Currently the most popular preconditioning method is that based on partial or incomplete Choleski factorization of the K matrix. In this method a lower triangular matrix L is constructed by considering only elements in a limited . band around the main diagonal of matrix K. Denoting this band by m, the incomplete Choleski decomposition algorithm is as follows:
For i = 1, n For j = 1, i-I If (i - j :$ m) then j-I )/ = ( leij - I:k=II ikI jk I jj Otherwise Iij = o Iij
end
Iii
= ~ leii - I:i-:,11 Itk
end With this lower triangular L matrix, the vector h in the PCG algorithm is established using forward elimination and backward substitution as follows:
SOLUTION OF LINEAR EQUATIONS
Ph.
=g
:::::::>
LLT/z
=g
Forward elimination for intermediate variablesy: Ly Back substitution for
h: LTh
=g
=y
Example 1.16 Find the solution of the following system of equations using the PCG method with incomplete Choleski decomposition:
[ ~ 1; 6 3
~ -1~][~~] = [~]
2 -10
11 -5
x3 x4
-5 21
7 8
The solution using a band m = 2 (two terms in each row around the diagonal) for generating the preconditioning matrix is as follows: Lower triangular factor of preconditioning matrix,
= (1.77636 X 10- 15, -1.77636 X 10- 15,8.88178 X 10- 16,4.44089 X 10- 151 Solving LLT w(3) = g(3), we have ",(3) = (1.03298 X 10- 16, 1.10676 X 10- 17, 1.35579 X 10- 16,2.49022 X 10- 16 ) r(3) = g(3)T W(3) = 1.39013 X 10- 30 /3(3) = r(3)/r(2) = 4.77262 x 10- 31 h(3)
=
_g(3)
+ /3(2)h(2)
= (-1.03298
X
10- 16 , -1.10676 X 10- 17, -1.35579 X 10- 16 , -2.49022 x 10- 16 )
The solution converged after three iterations: g r
= (1.77636 X 10- 15, -1.77636 X 10- 15,8.88178 X 10- 16,4.44089 x 10- 15 ) =gTg = 1.39013' X 10- 30
In some finite element modeling situations it becomes necessary to introduce constraints between several different degrees of freedoms. Such constraints are known as multipoint constraints and in general are expressed as follows:
c n d 1 + c12d2 + C21d 1
+ c 22d2 +
+ c 1n dn
= ql
+ c 2n dn =
q2
where cij ' i, j = 1,2, ... , and qi' i = 1, 2, ... , are specified constants and d.; i = 1, 2, ... , are the nodal degrees of freedom. In matrix form the constraints equations can be expressed as follows: Cd=q
where with m constraints C is an m X n matrix and q is an m x 1 vector. A common situation is modeling an inclined roller support. As is clear from Figure 1.22, in this case the displacement component normal to the incline is zero but neither
MULTIPOINT CONSTRAINTS
y
2
2
------------ x Figure 1.22. Inclined roller support
its horizontal or vertical displacement is zero, and hence this boundary condition cannot be incorporated using techniques discussed earlier. With the support malcing an angle a with the horizontal, the component of u l normal to the support is u l sin(a) and that of VI is vI cos(a). Thus we must solve the global system of equations with the following multipoint constraint between the degrees of freedom at this node: UI
sin(a) + vI cos (a)
=0
Another common use of multipoint constraints is in creating valid meshes when transitioning from a coarse to a fine mesh. As discussed earlier, the mesh consisting of three four-node quadrilateral elements in Figure 1.23 is invalid because node 4, which forms a corner of elements 2 and 3, is not attached to one of the four corners of element 1. A valid mesh can be obtained if the displacements at node 4 are related to the displacements at nodes 3 and 5. For the case when the node 4 is located at the midway point on line 3-5 the mesh compatibility can be achieved by defining two multipoint constraints defining displacements at node 4 as averages of the displacements at nodes 3 and 5. Thus we solve the system of equations with..the two following multipoint constraints: u4
U 3 + Us = --2===? u3 -
v3 +vs v4 = --2-
===?
y
2u 4 + Us
=0
v3 - 2v 4 + vs-= 0 2
5
8
---------x Figure 1.23. Multipoint constraints to create valid mesh in transition region
73
74
FINITEELEMENTMETHOD:THE BIG PICTURE
Figure 1.24. Rigid element in a finite element mesh
As a final example that requires use of multipoint constraints, we consider a situation in which some portions of a finite element model are much stiffer as compared to the rest. To avoid numerical difficulties in solving the resulting system of equations, it generally is more efficient to treat the stiff regions as completely rigid. Figure 1.24 shows a simple example in which the dark shaded triangle is assumed to be rigid. The nodes at the comers of this triangle must displace in such a manner that the original shape is maintained. A planar rigid body has only three degrees offreedom (two translations and a planar rotation). Thus .the six degrees of freedom of the element must be related by three constraint equations. For a triangle the three constraint equations can be written by recognizing the fact that the lengths of the sides of the triangle must remain unchanged between the initial configuration and the deformed configuration. In the initial configuration coordinates of the nodes of the rigid triangle are (x" YI)' (x2, Y2)' and (x3, Y3 ). In the deformed configuration these coordinates will be (XI + UI' Yl + VI)' (x2 + £1 2, Y2 + v 2), and (x3 + u 3' Y3 + v3)' Equating lengths of sides obtained from these coordinates gives us the following three equations:
/ (-£1 1+£12 (-£1 2
+ U3 -
(U I - U3
+ y2f =
(X 2 _x l)2
+ (Y2 -
+ x 3)2 + (-V 2 + V3 - Y2 + y3f =
(X 3 - x 2)2
+ (Y3 - y2f
(XI
+ (YI
XI +X2)2 X2
+ XI
+ (-VI
- x 3)2
+ (VI
+V2 -YI
- V3
+ YI
- Y3)2
=
X 3)2
YI)2.
- Y3)2
Expanding these equations and for small displacements neglecting the displacement squared terms (uI, U I U2'" .), we get the following equations: 2u 1x I - 2u 2x I - 2U IX 2
+ 2U 2X2 + 2v IYI
2U 2X2 - 2U 3X2 - 2U 2X3
+ 2U 3X3 + 2V2Y2 -
2V 3Y2 - 2V 2Y3
+ 2V 3Y3 = 0
2u lxI - 2u 3x I - 2U IX3
+ 2U 3X3 + 2v I yi -
2v3YI - 2V IY3
+ 2V 3Y3 = 0
- 2v2YI - 2V IY2
+ 2V 2Y2 = 0 ,
-Choosing £II' VI' and u2 as the three independent degrees of freedom (usually called the master degrees of freedom), we can solve these equations for the remaining three degrees of freedom (called the slave degrees of freedom). The resulting equations are written in the
MULTIPOINT CONSTRAINTS
uz=l, Ul=VI=a 3
Figure 1.25. Rigid-body modes for a plane triangle
form of the following multipoint constraints:
or
z z - - - U l + VI + - - - U z YI - Yz YI - Yz Y3 -Yz Yl -Y3 ---Ul + ---Uz YI -Yz Yl - Yz Xl -X
X -X _I_ _ 3U
YI - Yz:
X
-Xl
+ v + X_3_-X _ 1 U? I
Yl - Yz -
a
- Vz
=
- U3
=a
-
V 3
=
a
The first column of this transformation matrix represents a rigid-body mode that sets u l = 1 and vI =Uz = a.Similarly, the second column represents a mode with VI = 1, ul = Uz = a and the third column U z = 1, u l = vI = a. These three rigid-body modes are shown in Figure 1.25. An arbitrary rigid zone in a finite element model can be handled by breaking the shape into rigid triangles and then using the constraints for each of the triangles. The idea can also be extended to three-dimensional tetrahedrals each consisting of four triangular faces.
1.6.1
Solution Using Lagrange Multipliers
In the presence of multipoint constraints, the solution of global equations obviously becomes complicated. The minimization form introduced in the previous section allows us to use standard optimization techniques for constrained problems. Thus we set up the following optimization problem for solution of nodal degrees of freedom: Findd to Minimize f = !dT Kd - d T R Subject to Cd - q = 0
75
76
FINITEELEMENT METHOD:THE BIG PICTURE
4
3
4
2 5
o
p
-3
o
(m)
5
Figure1.26. Five-bar truss with inclined support
A standard optimization technique for solution of this problem is to introduce m Lagrange multipliers Ai' one for each equality constraint, and define an equivalent unconstrained problem as follows (for details see Bhatti [95]): Find d and A to Minimize L = !dT[(d -dTR + AT(Cd - q) where L is known as the Lagrangian. The necessary conditions for a minimum of the Lagrangian are that its derivatives with respect to d, and Ai be zero. Carrying out differentiations, the resulting equations are as follows:
aL ad . aL / -=O~Cd-q=O aA
-=O~[(d-R+CTA=O
Writing the two sets of equations together, the complete system of linear equations can be expressed as follows:
This system of linear equations can be solved using any of the methods discussed previously in this chapter. For structural problems a Lagrange multiplier can be interpreted as the force necessary to apply the specified constraint.
Example 1.17 Truss with an Inclined Support Consider the five-bar pin-jointed structure shown in Figure 1.26. All members have the same cross-sectional area and are of the same material, E = 70 GPa, and A = 10-3 m2 . The load P = 20 leN. The dimensions in meters are shown in the figure. For numerical calculations, the N . mm units are convenient. The displacements will be in mm and the stresses in MPa. The complete computations are as follows:
MULTIPOINT CONSTRAINTS
specified Nodal Loads Node
dof
Value
3
u3
20000.
Equations for Element 1
E
= 70000; A = 1000
Element Node
Global Node Number
x
Y
1 2
1 3
5000.
-3000.
Xl = 0;
YI = 0;
Xz = 5000.;
Yz = -3000.
L = ~ (xz - xI)z + (yz - YI)Z = 5830.95 Direction cosines:
1S
= Xz -xl = 0.857493', L
m
S
= Yz L-YI = -0514496 .
Substituting into the truss element equations, we get
-8827.13 5296.28](U I ] 5296.28 -3177.77 vI 5296.28 8827.13 -5296.28 u3 ( -8827.13 . 5296.28 -3177.77 -5296.28 3177.77 v3
(0.]
8827.13 -5296.28
-5296.28
311'J.77
O. O. O.
_ -
The element contributes tQ{I, 2, 5, 61 global degrees offreedom. Adding element equations into appropriate locations, we have
Then we remove (3,4) rows and columns. After adjusting for essential boundary conditions, we have 17654.3 0 -8827.13 5296.28 -8827.13 -5296.28
0 29688.9 5296.28 -3177.77 -5296.28 -3177.77
-8827.13 5296.28 8827.13 -5296.28 0 0
-8827.13 -5296.28 0 0 22827.1 5296.28
5296.28 -3177.77 -5296.28 14844.4 0 -11666.7
ul
-5296.28 -3177.77 0 -11666.7 5296.28 14844.4
0 0 20000. 0 0 0
vI
u3 v3
u4 v4
The multipoint constraint due to inclined support at node 1 is u l sin(7f/6)+ VI cos(7f/6) = O. The augmented global equations with the Lagrange multiplier are as follows: 17654.3
Solving the final system of global equations, we get
= 5.14286, vI = -2.96923, u3 = 16.8629, u4 = -1.42857, v4 = 11.7594, it = 80000.}
{U I
v3
= 12.788,
Solution for Element 1 Nodal coordinates Element Node
Global Node Number
x
y
1
1
0
o
2
3
5000.
-3000.
=
0 0 20000. 0 0 0 0
MULTIPOINT CONSTRAINTS
XI
== 0;
Yj
== 0;
X2
L = ~ (x 2 - X I)2 + (Y2 Direction cosines:
Is =
- YI)2
T-x
X
= 5000.;
Y2
zi:
-3000.
= 5830.95
= 0.857493;
In
s
= Y2 L-Yj = -0514496 .
Global-to-local transformation matrix:
T
= (0.857493 o
-0.514496 0
0 0.857493
0 ) -0.514496
Element nodal displacements in global coordinates:
d
(5.14286] = Ut] vI = -2.96923
( u3
v3
16.8629 12.788
Element nodal displacements in local coordinates: d I
= Td == (5.93762) 7,88048
E = 70000; A = 1000 Axial strain, E = (d2 - dj)IL = 0.000333197 Axial stress, o: = EE = 23.3238 Axial force, erA = 23323.8 In a similar manner, we can compute the solutions over the remaining elements: .Stress 1 - .2 3 4 5
23.3238 23.3238· 69.282 -20. -12.
Axial Force 23323.8 23323.8 69282. -20000. -12000.
1.6.2 Solution Using Penalty Function The Lagrange multiplier method of imposing constraints has two drawbacks. First, it requires adding new rows and columns to the global system of equations that for large systems may be inefficient. Second, the resulting system has zeros on the diagonal corresponding to the constraint equations. Some simple equation. solvers that assume nonzero diagonal terms may not work for this system. Another standard technique of imposing constraints, the so-called penalty function approach, does not have these two drawbacks. In this method a large penalty number f1 is chosen and the equivalent unconstrained problem is defined as follows (for details see Bhatti [95]):
79
80
FINITE ELEMENTMETHOD: THE BIG PICTURE
Findd to Minimize ¢
= ~dTKd -
dTR + ~fJ.(Cd - ql (Cd - q)
With fJ. being large, the minimization process forces the constraints to be satisfied. The necessary conditions for the minimum result in the following system of equations:
Rearranging terms, the system of linear equations can be expressed as follows:
This system of linear equations can be solved using any of the methods discussed previously in this section. The performance of the method depends on the value chosen for the penalty parameter u. Large values, say of the order of fJ. = 10 10 , give accurate solutions; however, the resulting system of equations may be ill-conditioned. IT fJ. values are small as compared to other terms in the global equations, the solution will not satisfy the constraints very accurately. A general rule of thumb is to set fJ. equal to 105 times the largest number in the global K matrix. Example 1.18 Truss Supporting a Rigid Plate A plane truss is designed to support a rigid triangular plate as shown in Figure 1.27. All members have the same cross-sectional area A = 1 in2 and are of the same material, E = 29,000 ksi. The load P = 20 kips. The dimensions in ft are shown in the figure. Note there is no connection between the diagonal members where they cross each other. The model consists of five nodes and thus the global system of equations before bound/ ' ary conditions will be 10 x 10. The equations for the six truss elements are written as in the previous examples and assembled in the usual manner to give the following system of equations:
0 0 142.693 -36.8215 -46.0268 36.8215 -96.6667 O. 0 0
-96.6667 0 -46.0268 36.8215 142.693 -36.8215 O. O. 0 0
0 '-120.833 -36.8215 150.29 36.8215 -29.4572 O. O. 0 0
0 0 36.8215 -29.4572 -36.8215 150.29 O. -120.833 0 0
-46.0268 -36.8215 -96.6667 O. O. : O. 142.693 36.8215 0 0
-36.8215 -29.4572 O. O.
0 0 0 0 0 0 0 0 0 0
o. -120.833 36.8215 150.29 0 0
81 0 0 0 0 0 0 0 0 0 0
"1 vI "2 v2
"3 v3
"4 v4
"5 v5
The essential' boundary conditions at node 1 (u l = vI = 0) are incorporated by removing the corresponding rows and columns in the usual way. Node 3 also has zero vertical displacement. However! since this node is connected to the rigid plate as well, the boundary condition v3 = 0 will be imposed later as part of the multipoint constraints. Removing the first two rows and columns, the global system of equations is as follows: .
Kd=R
~
142.693 -36.8215 -46.0268 36.8215 -96.6667
-36.8215 150.29 36.8215 -29.4572
-46.0268 36.8215 142.693 -36.8215
O.
O. O.
O. O.
0 0
0 0
0 0
36.8215 -29.4572 -36.8215 150.29
O. -120.833 0 0
O. O. O.
-96.6667
O. O. O. 142.693 36.8215 0 0
-120.833 36.8215 150.29 0 0
The rigid plate is connected between nodes 3, 5, and 4. Treating u3 ' dent degrees offreedom, the multipoint constraints are as follows:
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0
u2 "2
O. O. O. O. O.
u3 v3 u4 v4
-20.
Us Vs
-20.
"s- and Us as indepen-
Expanding and rearranging, we have
To this list we must also add the roller support constraint that "3 = O. Thus the complete set of constraint equations, expanded to include all degrees of freedom present in the global equations, are
O.
0 0 0 0 0 0 0 -20 0 -20
82
FINITEELEMENTMETHOD:THE BIG PICTURE
Uz
[0 oo
.
Cd=q~ ~ o o
-~
-1
0 0 0
-1
o 1
0 0 ~ 1 0 -1 0 1 0 0
0
0
Vz
~] t1 =m Us
Vs
To use the penalty function approach, we choose the penalty parameter j1 equal to lOs times the largest number in the global K matrix: j1
= 150.29 X lOs = 1.5029 X 107
Incorporating the constraints into the global equations with this value of u, the final system of equations is as follows: (K+pCTCjd=R+pCT q=
liP
0.00142693 -0.000368215 -0.000460268 0.000368215 -0.000966667 O. 0 0
-0.000368215 0.0015029 0.000368215 -0.000294572 O.
o.
0 0
-0.000460268 0.000368215 -234.827 -187.863 O. O. 234.829 187.863
0.000368215 -0.000294572 -187.863 -450.87 O. 150.289 187.863 150.29
O. O. O. 150.289 0.000368215 -150.289 0 0
-0.000966667 O. O. O. -150.289 0.000368215 150.29 0
0 0 234.829 187.863 150.29 0 -385.119 -187.863
0 0 187.863 150.29 0 0 -187.863 -150.29
O. O.
"2 "2
o.
1/3 "3
"4 "4 1/5 "5
O. O. -20. O. -20.
Solving this system of linear equations, we get (U z U4
= 0.172845,
Vz
= 0.076446, u3 =-0.139174, v3 = 3.99227 X 10- 6,
= 0.292292, v4
6.0391~·X 10-6,
Us
= 0.29229,
Vs
= -0.539324)
Substituting these values into the constraint equations, we can see that the constraints are reasonably satisfied: 6
1.33076 x 10- 6 ] Cd = 1.66345 x'1O6 [ 2.0469 X 103.99227 X 1O~6
",
[0] 0 0 0
Knowing the nodal values, the element solutions can be computed in the usual manner:
1 2 3 4 5 6
Strain
Stress
Axial Force
0.000318525 -0.000463913 0.000594098 0.000398156 -0.000509888 8.52877 x 10-9
Using the Lagrange multiplier method, the solution is obtained as follows: Augmented system of equations: 142.693 -36.8215 -46.0268 36.8215 -96.6667 O. 0 0 0 0 0 0
-'-36.8215 150.29 36.8215 -29.4572 O. O. 0 0 0 0 0 0
= 0.172849, dz = 0.0764461, d3 = -'-0.139174, d4 = -4.68418 X lO- IS, ds = 0.292296, d6 = -1.62088 X 10- 17, d7 = 0.292296, d s = -0.539337, AI = -20., Az = -25., A3 = -30.7628, A4 = -60.)
(d l
1.7
UNITS
It is important to use a consistent system of units during finite element calculations. When using the U.S. customary system of units, it is convenient to use the pound-inch (PI) units. Thus, the nodal coordinates will be entered in inches, the modulus of elasticity in pounds per square inch (psi), mass density in pounds per inch cubed (lb,/in3) , thermal conductivity in British thermal units per hour-inch-degrees Fahrenheit [Btu/(hr . in . OF)], and coefficient of thermal expansion in inch per inch per degrees Fahrenheit (in/in/°F). When using the International System of Units, it is recommended to use the newton-millimeter (N . mm) units. Thus the nodal coordinates will be entered in millimeters, the modulus of elasticity in megapascals (MPa), mass density in metric tonnes per millimeter cubed (mt/mm"), thermal conductivity in watts per millimeter-degrees Celsius [W/(mm· °C)),"and coefficient of thermal expansion in millimeters per millimeter per degrees Celsius (mm/mm/°C). Typical numerical values for these quantities for concrete, steel, and aluminum are as follows: Concrete (medium strength) Quantity Modulus of elasticity Mass density Thermal conductivity Coefficient of thermal expansion
PI System 4.64 X 106 psi 2.24 x 1O-4 1b,/ in3 0.048 Btu/(hr . in . OF) 6.1 x 10-6 in/in/T'
N -rnm System . 0.32
X
lOs MPa
'2.4 x 10-9 mt/rnm'' 0.001 W/(mm· 0c)
i 1 x 1O-6 mm/mm/' C
0 0 0 1 0 0 0 0 0 0 0 0
dl
~
d3 d4
ds d6 d7 dB it! A2 1\3
A4
O. O. O. O. O. -20. O. -20. 0 0 0 0
84
FINITE ELEMENT METHOD: THE BIG PICTURE
Steel (carbon and alloys):
PI System
N· mm system
30'x psi 7.32 x 1O-4 1b,/ in3 2.07 Btu/(hr· in- OF) 6.5 x 10- 6 in/in/T'
2.1 x MPa 7.83 x 10-9 mt/mm'' 0.043 W/(mm· 0C) 12 X 10-6 mm!mm!°C
PI System
N -rnm System
10.4 x psi 2.62 x 1O-4 1bll/ in3 10.11 Btu/(hr· in . OF) 12.8 x 10- 6 in/in/OF
0.72 x MPa 2.8 x 10- 9 mt/mnr' 0.21 W/(mm' °C) 23 X 10- 6 mm!mm!°C
Quantity Modulus of elasticity Mass density Thermal conductivity Coefficient of thermal expansion Aluminum alloys: Quantity Modulus of elasticity Mass density Thermal conductivity Coefficient of thermal expansion
Commonly used conversion factors:
Length Force or weight Weight to mass Weight to mass. Mass Mass density Power Temperature Coefficient of thermal expansion Thermal conductivity Convection coefficient Modulus of elasticity or stress
Given in
Multiply by
To get
in Ib! Ib!
mm
Ibm Ib,/in3 Btu/hr OF in/in/OF
25.4 4.4484 1/386.4 1.02 x 10- 4 0.17524 1.0694 x 10- 5 0.2931 ("F - 32)/1.8 1.8
Ibm mt mt mt/mm" W °C mm!mm!°C
Btu/(hr . in . ~F) Btu/(hr . in2 • OF) psi
0.02077 6.411 x 10-3 6.895 x 10-3
W/(mm·°C) W/(mm2 • 0C) MPa
/N
N
PROBLEMS Element Equations
1.1 A triangular finite element for a 2D heat flow problem is shown in Figure 1.28. Write the finite element equations for this element. Assume that the heat conduction
PROBLEMS
coefficient is 1.4 W/m· °C. The convection heat loss takes place along side 2-3 of the element. The average convection heat transfer coefficient is 20W/m' °C and the surrounding air temperature is 2YC. There is no heat generation inside the element. y(m)
0.2
3
0.15 0.1
- - - - - - - x(m)
o
0.05 0.1 0.15 0.2
Figure 1.28. Triangular element
1.2
A triangular finite element for a plane stress problem is shown in Figure 1.29. Write the finite element equations for this element. Assume E = 20 X 106 Nzcm", v = 0.3, thickness = 2 em, and a uniform pressure of 300 N/cm 2 normal to side 1-3 of the element.
y(cm) 0.1
3
y(m)
0.1
3
o o o Figure 1.29. Triangular element
1.3
-0.05 0.01 Figure 1.30. Triangular element
A triangular finite element for a 2D heat flow problem is shown in Figure 1.30. Write the finite element equations for this element. Assume that the heat conduction coefficient is 1.17 W/m' "C. The convection heat loss takes place along side 3-1 of the element. The average convection heat transfer coefficient is 17 W/m' °C and the surrounding air temperature is 32°C. There is no heat generation inside the element.
1.4 Consider two-dimensional steady-state heat flow in a V-grooved solid body with a cross section as shown in Figure 1.31. A hot liquid flowing through the groove maintains the surface at a temperature of 170°F. The thermal conductivity of the material is k = 0.04 Btu/hr- in- "F, The bottom surface is insulated. The sides have a convection coefficient of h = '0.03 Btu/hr . in 2 • OF with 'the outside temperature of 70°F. Determine the temperature distribution in the solid. Taking advantage of symmetry and using only two triangular elements, the model is as shown in the figure. Develop finite element equations for element 1.
85
86
FINITE ELEMENT METHOD:THE BIG PICTURE
4
Insulated Figure 1.31. V-grooved solid
1.5
Develop finite element equations for element 2 of the model shown in Figure 1.31.
1.6
The cross section of a long concrete dam (thermal conductivity = 0.6 W/m . DC) with base = 3 m and height = 4 m is shown in Figure 1.32. The face of the dam is exposed to heat flux qo = 800 W/m2 from sun. The vertical face is subjected to convection by water at 15DC with convection heat transfer coefficient 150 W/m2.DC. It is proposed to determine the temperature distribution in the dam using only one triangular element. Set up the system of finite element equations.
Figure 1.32. Concrete dam
Assembly of Element Equations
1.7
A three-bar truss is shown in Figure 1.33. Write the element equations and carry out the assembly of the element matrices to form the global system of equations. The area of cross section of each element is 500 mm 2 and E = 210 GPa. Use N . mm units in your calculations.
PROBLEMS
10 leN 0.5 0.4 0.3 0.2 0.1 0 (m)
0
0.1 0.2 0.3 0.4 0.5 0.6 Figure 1.33. Plane truss
1.8
A three-bar truss is shown in Figure 1.34. Write the element equations and cany out the assembly of the element matrices to form the global J( matrix and the R vector. The area of cross section of each element is 500 mm 2 and E = 210 GPa. Use N . mm units in your calculations. 0.6 0.5 0.4
45° 20 leN
0.3 0.2 0.1
2
o o
0.1 0.2 0.3 0.4 0.5
Figure 1.34. Plane truss
1.9
A finite element model employs triangular plane stress elements and consists of only six nodes. Each node has two degrees of freedom. The first element is connected to the model through nodes 1,4, and 6 and has the following coefficient matrix and the right-hand-side vector: 246 198
Carry out the assembly of the element matrices to form the global K matrix and the R vector. Note that this is the first element that is being assembled, and hence the global matrices are all zeros prior to assembly of this element.
1.10
Form the global system of equations for the two-element finite element model shown in Figure 1.35. The equations for the individual elements are as follows:
Handling Essential Boundary Conditions and Solution for Nodal Unknowns I
1.11
In Problem 1.7 you were asked to obtain a global system of equations for a plane truss. Continue further with the analysis and obtain the reduced system of equations after adjusting for specified displacement boundary conditions. Solve the resulting system for the unknown nodal displacements. Compute reactions at the supports and verify that your solution satisfies overall equilibrium. .
1.12 The global K matrix and the R vector for a finite element system are as follows:
63993.2 79991.4 0 99989.3 0 79991.4 D O. 0 O. 0 0 K= 0 0 0 0 0 0 -63993.2 -79991.4 O. -79991.4 -99989.3 O. RT
=(0.
O.
O.
o.
O.
O.
-63993.2 -79991.4 0 0 0 -79991.4 -99989.3 0 0 0 O. O. 0 O. 0 O. -210000. 210000. 0 0 239682. -59920.6 -239682; 59920.6 0 14980.1 -59920.6 59920.6 -14980.1 0 59920.6 303675. 20070.9 O. -239682. -210000. 20070.9 324969. 59920.6 -14980.1 14142.1 -24142.1)
89
PROBLEMS
Obtain the reduced system of equations after imposing the following essential boundary conditions: dl
= -0.1;
dz = 0.2;
d3
= d4 = d s = d 6 = 0
Solve the final system of equations for the remaining two nodal unknowns.
1.13
After assembly a finite element model yields the following global system of equations:
Get the reduced system of equations after imposing essential boundary conditions d l = -1, d z = 0, and d4 = 4. Solve the reduced system for remaining nodal unknowns.
1.14
A typical truss to support a highway is as shown in Figure 1.36. Because of a construction error, the support a~de 5 rom too tall. The contractor is considering using jack hammers to force the truss to fit it into place. You are asked to conduct a finite element analysis to make sure that the stresses in the truss elements will remain within allowable limits under the given loading and the forced fit. A colleague of yours has started the analysis already and has obtained the following global stiffness matrix after assembly of the first eight elements (N . rom units are used): 7560. -1920. -5000. 0 0 0 0 0 -2560. 1920. 0 0
The following nodal temperatures are obtained for the heat flow problem in the Vgrooved solid shown in Figure 1.31. Compute the temperature distribution and its x and y derivatives over each element. By comparing the temperature derivatives over the two elements, comment on the quality of the solution. Node
Temperature
1 2 3 /4
99.4118 170 28.8235 170
The correct nodal displacements, in mm, for the three-bar in Problem 1.7 are as follows: u
v
1
0
2
0
o o
3 4
0 0.0506837
-0.0736988
o
Compute axial strains, axial stresses, and axial forces in each element. Using the computed axial forces, draw a free-body diagram of all forces acting at node 4. By summing forces in the horizontal and vertical directions, show that the joint equilibrium is satisfied. 1.17
Determine joint deflections, stresses in each member, and support reactions for the plane truss shown in Figure 1.37. The members have a cross-sectional area of 8 cm2 and are made of steel (E == 200 GPa). Show all calculations.
PROBLEMS
1.5
10kN
0.5
o o
(m)
1.5
0.5 Figure 1.37. Plane truss
1.18 Determine the temperature at the centroid of the concrete dam cross section shown in Figure 1.32, assuming the temperatures at the nodes are 10, 20, and 30°C, respectively. Solution of Linear Equations
1.19 Solve the following system of equations using Choleski decomposition:
-; -~
[
~
oo
04· 9 0 0 .. -1.5 000
-1.5 1.25 0.5
000
-2 [2 ] 8
x
2 ][Xl] x =
0.5 3.25
3
x4
0.5
X
-3.25
s
1.20 Solve the following system of equations using Choleski decomposition:
1.21 Factor the following matrix using Choleski decomposition: 2 -1
o o o
000 -1 0 o 0 0 2-1 O' 0 -1 3 -1 3 -1 0 0-1 2-1 0 0 -1
000
o -1
2
91
92
FINITE ELEMENT METHOD: THE BIG PICTURE
Using the factored matrix, obtain solutions for the following right-hand sides: ~l
b(l)
1.22
=
~l
~l
0 0 0 0 0
b(2)
=
2 0 0 0 0
b(3)
2 0 0 0
=
~2
Factor the following matrix using Choleski decomposition:
Using the factored matrix, obtain solutions for the following right-hand sides:
0
-8
b(l)
-3 14 = 39 32 1 -45 56 1
,/
r
b(2)
15 43 43 27 = 12 14 59 79 -2 63
6 10
8 b(3)
12 7 = 0 29 19 -24 21
1.23
Solve the system of equations in Problem 1.19 using the basic conjugate gradient method.
1.24
Solve the system of equations in Problem 1.20 using the basic conjugate gradient method.
1.25 -Solve the system of equations with the given coefficient matrix and the first righthand side in Problem 1.21 using the basic conjugate gradient method. 1.26 Solve the system of equations with the given coefficient matrix and the first righthand side in Problem 1.22 using the basic conjugate gradient method. 1.27
Solve the system of equations in Problem 1.19 using the conjugate gradient method with Jacobi preconditioning.
PROBLEMS
1.28
Solve the system of equations in Problem 1.20 using the conjugate gradient method with Jacobi preconditioning.
1.29
Solve the system of equations with the given coefficient matrix and the first righthand side in Problem 1.21 using the conjugate gradient method with Jacobi preconditioning.
1.30
Solve the system of equations with the given coefficient matrix and the first righthand side in Problem 1.22 using the conjugate gradient method with Jacobi preconditioning. .
1.31
Solve the system of equations with the given coefficient matrix and the first righthand side in Problem 1.21 using the conjugate gradient method with incomplete Choleski preconditioning with 112 == 1. The solution should converge in one iteration. Why?
1.32
Solve the system of equations with the given coefficient matrix and the first righthand side in Problem 1.22 using the conjugate gradient method with incomplete Choleski preconditioning with 112 == 1.
Multipoint Constraints 1.33
A finite element system results in the following system of global equations:
2 -2 -2 6
0 4
049 [
o o
0-1.5 0 0
o
o
-2
28 ] 0.5 -3.25
-1.5 1.25 0.5
Using the Lagrange multiplier approach, solve for the nodal unknowns if the model requires the following multipoint constraint: . dl+d z == 1 1.34
A finite element system results in the following system of global equations:
Using the Lagrange multiplier approach, solve for the 'nodal unknowns if the model requires the following multipoint constraints:
93
94
FINITE ELEMENTMETHOD:THE BIG PICTURE
1.35
Solve Problem 1.33 using the penalty function approach.
1.36
Solve Problem 1.34 using the penalty function approach.
Computational Projects 1.37
Using a finer mesh and the available finite element software, determine the temperature distribution through a solid with a V-groove as shown in Figure 1.31. Take advantage of symmetry and model only half of the domain.
1.38 Determine the temperature distribution through a solid with a circular groove as shown in Figure 1.38. ,A hot fluid flowing through the groove keeps the temperature of the circular surface at 150°C. The bottom surface of the solid is in contact with a coolant that maintains it at DoC. The top and side surfaces are well insulated. The solid is made of two different materials, with thermal conductivities k1 =: lOW/rn- "C and k2 =: 20 Wlm . DC. Take advantage of symmetry and model only half of the domain.
f
10 m
~I Figure 1.38. Solid with a circular groove
1.39
Heating wires are embedded in a concrete slab, as shown in cross section in Figure 1.39, to keep snow from accumulating on the slab. In a proposed design wires are placed at 2 em from the top and at 4-cm intervals. The heat generated by each wire is 10 Wlm length. The bottom of the slab is insulated. T):J.e top is exposed to a convection heat loss with a convection coefficient of h =: 30 W1m2 • DC. The thermal conductivity of concrete is 1.2 Wlm . DC. Determine the slab surface temperature when the air temperature is -SOC. Note that, because of symmetry, we need to model
T
6cm
1 Figure 1.39. Heating wires embedded in concrete
95
PROBLEMS
only the 2 ern x 6 em dark shaded area. There will be no heat flow across the sides of this area. The heating wire represents a concentrated point heat source. 1.40
Steps of a staircase are supported by two identical trusses made of 2-in-nominaldiameter pipes (outside diameter = 2.375 in, wall thickness =. 0.154 in). One side truss is shown in Figure 1.40. There are 12 steps each with 7.5 in rise and 11.5 in run. The load from the steps is assumed to be applied as concentrated downward loads of 100 lb at each nodal point. Determine nodal deflections, stresses in each member, and support reactions. Verify computer results by performing the following calculations by hand: (a) Use computed reaction forces and applied external forces to check over all equilibrium of forces. (b) Isolate a typical node in the truss and draw a free-body diagram of the forces acting at that node. Verify that the equilibrium is satisfied at the node. (c) Pass a vertical section through the truss. Draw the free body diagram of the truss on the right-hand section. Verify that the section is in equilibrium.
Figure 1.40. Staircase truss
1.41
The lower portion of an aluminum step bracket, shown in Figure 1.41, is subjected to a uniform pressure 'of 20 Nzrnrrr'. The left end is fixed and the right end is free. The dimensions of the bracket are thickness = 3 rum, L = 150 rum, b = 10 rum,
Figure 1.41. Step bracket
E::o 'l qi e.G("l 5;
96
FINITE ELEMENTMETHOD: THE BIG PICTURE
X ~( ..
M~
~ and h = 24 mm. Fillets of 3 mm radius are provided at the two reentrant comers. The material properties are E = 70,000 N/mm 2 and v = 0.3. Determine stresses in the bracket using a planar finite element model (using say Plane42 element of ANSYS). Compare solutions based on plane stress and plane strain assumptions. (The situation can also be modeled using bearn/frame elements, three-dimensional solid elements, and plate elements.)
~
'0~lJ ,t:J '.<. l.:J"
,.J
R ':f':(.0
I
1.42 The top surface of an S-shaped aluminum block, shown in Figure 1.42, is subjected to a uniform pressure of 20 N/mm2 . The bottom is fixed. The dimensions of the block are t = 3 mm, L = 15 mm, b = 10 mm, and h = 24 mm. The material properties are E =70,000N/mm 2 and v. = 0.3. Determine stresses in the block using a planar finite element model (using say Plane42 element of ANSYS). Compare solutions based on plane stress and plane strain assumptions. (The situation can also be modeled using bearn/frame elements, three-dimensional solid elements, and plate elements.)
Figure 1.42. S-shaped aluminum block
1.43
Design optimization. Aluminum fins are to be used to cool an integrated circuit (IC) chip as shown in Figure 1.43. On~ side of the chip is insulated. The chip dissipates electrical energy at a rate of 150 W1m. The thermal conductivity of aluminum is 170 W1m . K. The convection coefiicient of the fin surface is 55 W1m2 • K, and the ambient air temperature is 30·C.
ooling fms
C Chi sulatiEn Figure 1.43. Aluminum fins for cooling IC chip
PROBLEMS
Since the length of the fins is long, we can take a cross section and model as a plane problem. Furthermore, taking advantage of symmetry, we need to model half of the section, as shown in Figure 1.44. The bottom of the fin receives heat flux from the IC chip. Because of symmetry, no heat can flow across the left side. Convection heat loss takes place along the remaining exposed sides.
Figure'1.44. Planar model for aluminum fins for cooling
Ie chip
The dimensions d, w, and t are fixed as follows: d= 3mm;
w=3mm;
t=lmm
The design goal is to determine the optimum height h of the fins so that the temperature ofthe IC chip does not exceed 70 Assume a starting value of h = 5 mm. QC.
97
CHAPTER TWO
MATHEMATICAL FOUNDATION OF THE FINITE ELEMENT METHOD
From a mathematical point of view the finite element method is a special form of the well-known Galerkin and Rayleigh-Ritz methods for finding approximate solution of differential equations. In both methods the governing differential equation first is converted into an equivalent integral form. The Rayleigh-Ritz method employs calculus of variations to define an equivalent variational or energy functional. A function that minimizes this energy functional represents a solution of the governing differential equation. The Galerkin method uses a more direct approach. An approximate solution, with one or more unknown parameters, is chosen. In general, this lassumed solution will not satisfy the differential equation. The integral form represents the residual obtained by integrating the error over the solution domain. Employing a criteria adopted to minimize the residual gives equations for finding the unknown parameters. For most practical problems solutions of differential equations are required to satisfy not only the differential equation but also the specified boundary conditions at one or more points along the boundary of the solution domain. In both methods some of the boundary conditions must be satisfied explicitly by the assumed solutioris while others are satisfied implicitly through the minimization process. The boundary conditions are thus divided into two categories, essential and natural. The essential boundary conditions are those that must explicitly be satisfied while the natural boundary conditions are incorporated into the integral formulation. In general,therefore, the approximate solutions will not satisfy the natural boundary conditions exactly. The basic concepts will be explained with reference to the problem of axial deformation of bars. The derivation of the governing differential equation is considered in the next section. Approximate solutions using the classical form of Galerkin and Rayleigh-Ritz methods are presented. Finally the methods are cast into the form that is suitable for developing finite element equations. 98
AXIAL DEFORMATION OF BARS
2.·1
AXIAL DEFORMATION Of BARS
2.1.1 Differential Equation for Axial Deformations Consider a bar of any arbitrary cross section subjected to loads in the-axial direction only, as shown in Figure 2.1. The area of cross section is denoted by A and it could vary over the length of the bar. The modulus of elasticity is denoted by E. The bar may be subjected to a distributed axial load q(x, t) along its length. The load can vary over the bar length and may also be a function of time. The axial displacement is denoted by u(x, t). The governing differential equation can be written by considering equilibrium of a differential element as shown in Figure 2.2. Note the sign convention adopted in drawing the free-body diagram assumes that the tension in the bar is positive. The axial force at x is denoted by F. The axial force at x + dx is F + (aFlax) dx based on the Taylor series expansion. The acceleration is indicated by ii == 2 ul at 2 . From Newton's second law of motion a force, called the inertia force, is produced that is proportional to the mass of the element and acts in the direction opposite to the direction of motion. Denoting the mass density by p, the mass per unit length is ni = Ap dx and thus the inertia force is mii = Ap dx ii. The only other force acting on the element is the applied axial load q(x, t). Considering summation of forces in the x direction, we have
a
.. A(x)p dx u(x, t) + F
ax
.. aF = q.dx + F + aF ax dx =? Apu = q +
The axial force is related to the axial stress 0:.: as follows:
Figure 2.2. Forces acting on a differential element in an axially loaded bar
99
100
MATHEMATICAL FOUNDATION OF THE FINITE ELEMENTMETHOD
Assuming linear elastic material, axial stress is related to the axial strain modulus of elasticity. Thus
Ex
through the
Assuming small displacements, the axial strain is related to the first derivative of the axial displacement. Thus the axial force is related to displacement as follows:
F =AE
au ax
Substituting this into the equilibrium equation, the governing differential equation for axial deformation of bars is as follows:
au axa (AE ax ) +q =Apu.. This is a second-order partial differential equation. Since the equation involves secondorder derivatives with respect to both x and t, we need two boundary conditions and two initial conditions for a proper solution. The initial conditions are specified displacement and velocity along the bar at time t = 0: it(x,O) =
V
o
where U o and Vo are the specified values. The boundary conditions involve specification of displacement or its first derivative at the ends. Since the first derivative of displacement is related to the axial force, the derivative boundary condition is expressed in terms of the applied force. The possible boundary conditions at the right end of the bar x = XI are as follows:
/
or
XI -?
right end of the bar
where u; is a specified displacement and Pxf is a specified force. Both these quantities could be functions of time. Care must be exercised in assigning proper signs to force boundary condition terms. Applied forces are considered positive when they act in the positive coordinate direction. From the free-body diagram in Figure 2.2, it should be clear that, if a force is specified at the left end of a bar, then the appropriate force boundary condition must include a negative sign as follows: '~l
X
o -? left end of the bar
If the forces and displacements do not vary with time, we have a static analysis situation and the equilibrium equation is an ordinary second-order differential equation as follows:
d( dU) + q(x) = 0;
dx AE dx
Xo
< x < XI
AXIAL DEFORMATION OF BARS
with the boundary conditions of the form or or An integration of the second-order differential equation, if possible, would yield two arbi-
trary constants, and we would need two conditions to evaluate these constants. Thus at least two boundary conditions must be specified for a unique solution. On physical grounds, it is easy to see that we must specify displacement at least at one point along the bar to prevent the whole bar from moving as a rigid body. The second bouridary condition could be either of the displacement or the force type. Note that at the same point both a displacement and a force cannot be specified independently. The reason for this is that, if a displacement is specified, then the corresponding force represents a reaction at the point and ingeneral is
2.1.2 Exact Solutions of Some Axial Deformation Problems It is possible to directly integrate the differential equation to obtain exact solutions for the axial deformation problems in which the loading and the area of cross section are simple functions of x. Solutions for a uniform bar and a tapered bar subjected to linearly varying load are presented in this section. In later sections these exact solutions will be used to check the quality of the approximate solutions that are obtained using the Galerkin, the Rayleigh-Ritz, and the finite element methods. Axial Deformation of a Uniform Bar Consider a uniform bar fixed at one end and subjected to a static point load at the other end, as shown in Figure 2.3. The bar is also subjected to a linearly varying axial load q(x) = ex, where e is a given constant. The problem is described in terms of the following boundary value problem: O
u(O) = 0;
dx
=P .
'}---I»-P
L
Figure 2.3. Uniform axially loaded bar
£!>/
101
102
MATHEMATICAL FOUNDATION OF THE FINITE ELEMENTMETHOD
An exact solution of the problem can easily be obtained by integrating the differential equation twice and then using the boundary conditions to evaluate the resulting integration constants. Integrating both sides of the differential equation once, we get 2
EA du + ex dx 2
=C
.
I
where C1 is an integration constant. Integrating once again, we get
c;2
EAu(x) + (5
= C1x + C2
where C2 is another integration constant. Rearranging terms,
Using the boundary conditions £1(0)
= 0 ~ C2 = 0
and
Thus the exact solution of the problem is _ x(6P + 3eL 2 ) u(x - , 6EA
-
ex 2 )
!
Axial Deformation of a Tapered Bar Consider a tapered bar fixed at one end and subjected to a static point load at the other end, as shown in Figure 2.4. The bar is also subjected to a linearly varying axial load q(x) = ex, where e is a given constant. The problem is described in terms of the following boundary value problem: d ( EA(x) dU) dx dx + ex A(x)
=A o -
£1(0)
= 0;
= 0;
O
A -A LX _O_ _ L
EA L
dueL) _ -P t;lx
where A o is the area of cross section at x = 0 and A L is that at x =L. Integrating both sides of the differential equation once, we get Ao-AL)dU ex?_C +2 -- 1 E(A o - - -L- x dx
AXIAL DEFORMATION OF BARS
p
Figure 2.4. Tapered axially loaded bar
where C1 is an integration constant. Straightforward integration once again is not possible because of the presence of the x du/dx term. However, using more advanced techniques for solution of ordinary differential equations, it is possible to obtain the following general solution in terms of two integration constants C1 and C2 :
Using the boundary conditions, the integration constants are evaluated as follows:
The exact solution of the problem can now be written as follows:
1 u(x)=-
. 3
4(-1 + r) EA o
(L(c(-1+r)x(-2L+(-1+r)x)
+ 2(2P('-1 + :)2 + cL2(-2 + r)r) log[L] - 2(2P(-1 + r)2 + cL2(-2 + r)r) log[L + (-1 + r)xJ)) where r =AJAo' the ratio of areas at two ends of the bar: The only difference between this example and the previous one is that here the area of cross section of the bar varies linearly. The solution is much more difficult to obtain and the final solution expression is quite complicated as well. This example demonstrates how quickly analytical solutions become difficult to obtain and shows the importance of numerical methods for practical problems. As should be expected, by taking the limit as Ao -) AL == A, i.e., r -) 1, it can be shown that this solution reduces to the one for a uniform bar presented in the previous section. Also note that, when c = 0 (no distributed axial load), the axial force at any point in the bar is equal to P, which obviously should be the case from statics.
103
104
MATHEMATICAL FOUNDATION OF THE FINITEELEMENTMETHOD
u(x)
2
r=0.2 r=O.4
1.5
r=0.6 r=0.8 r=1.2 r=2 0.5 r=6 r=10
x 0.2
0.4
0.6
0.8
Figure 2.5. Solutions of tapered axially loaded bar subjected to end load
Solutions for several values of r are plotted in Figure 2.5 with the following numerical values:
2.2
A o = 1;
p= 1;
c = 0;
L= 1;
E=1
AXIAL DEFORMATION OF BARS USING GALER KIN METHOD
The governing differential equation for axial deformation of bars is the following secondorder differential equation: ./
d( dU)
dx AE dx + q
=:=
0;
where X o and XI are the coordinates of the ends of the bar. The primary unknown is the axial displacement u(x). Once the displacement is known, axial strain, stress, and force can be computed from the following relationships: . E
x
du
=-' dx'
F
=AD;:
At the ends either a displacement or an axial force can be specified. Thus the boundary conditions for the problem are of the following form: or or
. AXIAL DEFORMATION OF BARS USING GALERKIN METHOD
WhBre uxQ' P.tO' ... are appropriate specified values. Mathematically spealcing, for a secondorder differential equation either u is specified or its first derivative du/dx is specified. Both u and du/dx cannot have specified values at the same point.
2.2.1
Weak Form for Axial Deformations
In the Galerkin method we assume a general form of the solution. This assumed solution must contain some unknown parameters whose values are determined so that the error between the assumed solution and the exact solution is as small as possible. The assumed solution can be of any form. As an example, we will consider a solution in the form of a polynomial:
where ao' a l , ..• are the unknown parameters. In general, this assumed solution will not satisfy the differential equation for all values of x. When this assumed solution is substituted into the differential equation, the error in satisfying the governing differential equation is e(x)
;=
d( dft)
dx AE dx + q(x)
'* 0
The tilde (~) over u is used to emphasize that it is an assumed solution and not necessarily the same as the exact solution of the differential equation. In the following development we . will have to perform several mathematical operations, such as differentiation and integration, on the assumed solution. Carrying the tilde symbol through all these manipulations becomes tedious. Since we are dealing primarily with the approximate solutions, for notational convenience, we will drop the tilde and use only u instead to indicate an approximate solution. An occasional reference to the exact solution will be indicated by using the word exact explicitly.. The total error, called the residual, for the entire solution domain can be obtained by integrating e(x) over the domain. However, in a straight integration of e, the negative and positive errors at different points may cancel each other. To avoid error cancellation, e(x) is multiplied by a suitable weighting function and then integrated over the solution domain. Since there are n unknown parameters, we need n weighting functions to establish weighted residual equations as follows: i
=0, 1, ... , n
where wi(x) are suitable weighting functions. This form is known as the weak form. The differential equation is the strong form because it requires error term e(x) to vanish at every x. The integral form makes the total error go to zero but does not necessarily satisfy the governing differential equation for all values of x. In a method known as the least-squared weighted residual method the error term is squared to define the total squared error as follows:
105
106
MATHEMATICAL FOUNDATION OF THE FINITEELEMENTMETHOD
The necessary conditions for the minimum of the total squared error give n equations that can be solved for the unknown parameters: .i = 0, 1,... Thus in the least-squares method the weighting functions are the partial derivatives of the error term e(x). Least-squares weighting functions: i
=0, l, ... .n
A more popular method in the finite element applications is the Galerkin method. In this method, instead of taking partial derivatives of the error function, the weighting functions are defined as the partial derivatives of the assumed solution. Galerkin weighting functions:
i = 0,1, ... , n
Thus the Galerkin weighted residual method defines the following n equations to solve for the unknown parameters: i
= 0,1, ...
It turns out that for a large number of engineering applications the Galerkin method gives the same solution as another popular method, the Rayleigh-Ritz method, presented in a later section. Furthermore, since the least-squares method has no particular advantage over the Galerkin method for the kinds of problems discussed in this book, only the Galerkin method is presented in detail. So far in developing the residual we have considered error in satisfying the differential equation alone. A solution must also satisfy boundary conditions. In order to be able to introduce the boundary conditions into the weighted residual, we use mathematical manipulations involving integration by parts.
AXIAL DEFORMATION OF BARS USINGGALERKINMETHOD
'I'he integration-by-parts formula is used to rewrite an integral of a product of a derivative of a function, say f(x), and another function, say g(x), as follows:
Note that the integrand must involve the product of a function and the derivative of another function. The application of the formula produces two terms that are evaluated at the ends of integration domain and another integral in which the derivative shifts from one function to the other. For a second-order differential equation, we know that the boundary conditions specify either u or du/dx at the ends. Thus we are looking for a way of introducing these terms into the weighted residual. Since the integration-by-parts formula gives rise to boundary terms, it is precisely the tool that we need. Note that the highest derivative term in the residual e(x) involves a second derivative on u. Using integration by parts on this term will result in two terms evaluated at the integration limits that are nsed to incorporate the boundary conditions into the residual. . For the axial deformation problem, the weighted residual is
Note that, in general, A and E could be functions of x, and therefore, care must be taken when carrying out differentiation and integration. Also q(x) and w/(x) are arbitrary functions of x at this stage and cannot be taken out of the integral. Writing the two terms in the weighted residual as separate integrals, we have
L
LXI q(x)wi(x) dx. = 0; X I d(dU) . dx AE dx wi(x)dx +
Xo
i = 0,1, ...
Xo
The first integral contains the second-order derivative on u and is written exactly in the form to which integration by parts is applicable with f(x) = AE(du/dx) and g(x) = w/(x). Thus the application of integration by parts to the first integral gives du(x) du(xo) A(x/)E(x/) d/ wi(x/)-A(xo)E(xo)~w/(xo)-
(XI
du dw.
Jxo AE dx
dx l dx +
(XI
J
q(x)w/(x)dx=
Xo
°
The first two terms in the weak form give us a way to incorporate the specified force or derivative boundary conditions into the weak form. If a force PxO is applied at end xo' then du(xo) -A(xo)E(xo) ~ =
P.to:=;. Second term in the weak form: P.to w;Cxo)
If a force px / is applied at end xI' then A(x/)E(x/)
dL~~/) = PXI:=;. First term in the weak form: P.t Wi(x/) l
107
108
MATHEMATICAL FOUNDATION OF THE FINITE ELEMENTMETHOD
Thus, when a force is specified, the boundary condition can be naturally incorporated into the weak form. Hence this type of boundary condition is called a natural boundary condition (NBC). The situation is very different if the u is specified at one or both ends. The derivative du/dx at the corresponding point is unknown. (Physically AE du/dx at the point represents the unknown reaction.) The specified displacement boundary condition therefore cannot be incorporated into the weak form directly. The assumed solutions must satisfy this type of boundary condition explicitly, and thus such a boundary condition is called the essential boundary condition (EBC). An assumed solution that satisfies the EBC is known as an admissible solution. Since the weighting functions are partial derivatives of the assumed solution, with an admissible assumed solution, all weighting functions corresponding to the location of an EBC are zero. Therefore, the boundary term in the weak form vanishes at the point where anEBC is specified. Thus we must deal with the boundary terms as follows: Boundary Condition Term
Specified Value
Boundary Term in the Weak Form
Requirement on the Assumed Solution
'IYpe
-A(xo)E(xo)C!Ld;o)
P.to
P.towj(xo)
None
NBC
A(x/)E(x/) dL~~/)
Px,
PX,wj(x/)
None
NBC
u(xo)
uXo
None
Must satisfy
EBC
None
Must satisfy
EBC
!t(x/)
For structural problems, the weak form can be interpreted as the well-known principle of virtual displacements. To see this, assume we have specified force boundary conditions at the ends. Then the weak form can be Written as follows:
Rearranging terms, we have
If we interpret wj(x) as a virtual displacement, then the right-hand side is the virtual work done by the applied forces. The left-hand side is the total internal virtual work, since E du/dx == 0:;, is the axial stress and dw.rdx is axial virtual strain. Thus the weak form implies that when a bar is given a virtual displacement, then the external virtual work is equal to the total internal virtual work, which is a statement of the principle of virtual displacements. Since this principle is widely used in structural mechanics, it is one of the main reasons why the Galerkin weighted residual. method is more popular in developing finite element equations. Also recall that the virtual displacements are required to satisfy the displacement (i.e., essential) boundary conditions.
AXIAL DEFORMATION OF BARS USING GALERKIN METHOD
From the final weak form we note that another advantage of the integration by parts is that the order of the derivative on the terms remaining inside the integral sign is reduced by 1. Thus for a second-order problem the weak form involves only first-order derivatives. It may not appear to be a big deal here, but it has important consequences in developing simple finite elements for practical problems. It will be seen in later example that, when dealing with a fourth-order problem, integration by parts must be carried out twice to reduce the highest order derivative to 2.
a
2.2.2 Uniiorm Bar SUbjected to linearly Varying Axial load We now use the Galerkin method to find approximate solutions for a uniform bar (EA constant) fixed at one end and subjected to a static point load at the other end, as shown in Figure 2.3. Thebar is also subjected to a linearly varying axial load q(x) = ex, where e is a given constant. The problem is described in terms of the following boundary value problem:
O
u(O) = 0;
dx
=P
The following exact solution of the problem was obtained in Section 2.1:
_ x(6P + 3eL2
u(x) -
-
6EA
ex2) .
With the solution domain fromIf), L), the EA constant, q(x) = ex, the essential boundary condition u(O) = 0 => w(O) = 0, and the natural boundary condition EAu' (L) = P, the weak form specific to this problem is as follows: Pw;(L)
EBC:
+
lL (~AE~: ~:i
+ eXWi(X))dX = 0
=0
u(O)
This weak form can now be used to find a variety of approximate solutions to the problem. Linear Solution The simplest possible solution that we can assume is a linear polynomial. An approximate solution of the problem is obtained using the following starting solution: u(X) = a o +xa l
To satisfy the essential boundary condition, we must have u(O) = 0
==> ao + Oal
= 0 ==> ao = 0
Thus the admissible assumed solution is u(x) = xa l
giving
109
110
MATHEMATICAL FOUNDATION OF THE FINITE ELEMENTMETHOD
Substituting into the weak form, we have Pw;CL) +
) i Jo(L( -AE(a 1) dw dx + cxw dx = 0 j
There is only one unknown parameter left in the solution and therefore we need only one equation to find it. The Galerkin weighting function is
ax -1' - ,
aWl
Substituting this into the weak:form, we have PL + lL(-AEa l + c2)dx = 0,
Carrying out integration and simplifying, we get
Solving this equation for a p we get -CL2 -
3P
3EA
Thus a linear approximate solution for the problem is as follows: u(x)
Quadratic Solution nomial:
= ~r1: =
(cL 2 + 3P)x 3EA
A better solution can be obtained if we start with a quadratic poly-
To satisfy the essential boundary condition, we must have
Thus the admissible assumed solution is
Substituting into the weak form; we have
'AXIAL DEFORMATION OF BARS USINGGALERKIN METHOD
There are two unknown parameters left in the solution and therefore we need two equations to find them. We get these equations by using the two Galerkin weighting functions aWl
= I:
ax ' ow _ 2 =2.:c ax '
WI(O)
= 0;
W 2 (0)
= 0;
Substituting WI into the weak form, we have
Substituting w 2 into the weak form, we have
Solving the two equations, we have
-7cL2 -12P l2EA Thus a quadratic approximate solution is as follows: 1)I
2+
(x - a2A
x-- cL 2_ -7cL2-l2P (12P+cL(7L-3x))x x_ a l - 4EA x l2EA l2EA
Cubic Solution The exact solution of the problem is a cubic polynomial. As demonstrated below, the Galerkin method finds an exact solution if we start with a cubic polynomial:
To satisfy the essential boundary condition, we must have
Thus the admissible assumed solution is
Substituting into the weak form, we have
111
112
MATHEMATICAL FOUNDATION OF THE FINITEELEMENTMETHOD
There are three unknown parameters left, which we find by using the three Galerkin weighting functions into the weak form, giving the following equations:
au aa l au Wz= - =xz. aaz ' wI = - =x;
au W3 = - =x3 . ' aa3
aWl -1' ax - ,
WI(O) = 0;
awz _ 2x' ax - ,
Wz(O)
= 0;
wz(L) = LZ
aw =3xz; ax
W3(0)
= 0;
w3(L) = L3
_3
wI(L) = L
Substitute admissible solution and weights into the weak form and perform integrations to get the following: . Weight
Since this is the exact solution, trying any higher order polynomial will not make any difference. Carefully note the distinction between the two types of boundary conditions. To make the assumed solution admissible, we required it to satisfy only the displacement boundary condition [u(O) = 0]. We never explicitly require the assumed solution to satisfy the force boundary condition. We can easily see that for this example the linear and quadratic solutions in fact do not satisfy this boundary condition. Only the cubic solution satisfies this condition exactly.
The logic in using the terminology essential and natural boundary conditions should now be clear. Essential boundary conditions are those that must explicitly be satisfied by the assumed solution while the natural boundary conditions are only implicitly satisfied through the weak form. A solution that satisfies the differential equation and all boundary conditions is obviously the exact solution.
2.2.3
Tapered Bar SUbjected to linearly Varying Axial Load
Consider now the tapered bar fixed at one end and subjected to a static point load at the other end, as shown in Figure 2.4. The bar is also subjected to a linearly varying axial load q(x) : ex, where e is a given constant. The problem is described in terms of the following boundary value problem:
d( dU)
dx EA dx + ex : 0;
0
A(x) : A _ A o - A Lx: (L + (-1 + r)x)A o o L L u(O): 0;
ErA dueL) : P o dx
where A o is the area of cross section at x : 0, A L is that at x: L, and r : ALIA o. The weak form specific to this problem is the same as that for the uniform bar in the previous section, except that A is now a function of x:
.PwJL) + EBC:
dw. ) Jo(L(_-AE du dx d; + exwj(x) dx: 0 u(O): 0
Starting~ssumed solution: u(x) : ao + xa 1
Linear Solution
The admissible solution must satisfy EBC: EBC
Equation
u(O) ~ 0
ao : 0
Thus the admissible assumed solution is u(x) : xa 1• Weighting function
-7
{x}
Substitute into the wealeform and perform integrations to get: Weight
Equation 4/3
x
LP + eL
- (l/2)rEa 1A oL2 L
-
(l12)Ea 1A oL2
:
0
113
114
MATHEMATICAL FOUNDATION OF THE FINITEELEMENTMETHOD
Solving this equation,
a 1 -
_(CL3/3) - PL -(l/2)LEAo - (1/2)LrEA o
--:-:-:::-:-=-'::::-:---''-:-:--:-:-.-::-=-:-
Substituting into the admissible solution, we get the following solution of the problem:
2cxL2 + 6Px u(x)=----. 3rEA o + 3EA o Quadratic Solution
Starting assumed solution: u(x) = a2x2 + a1x + ao
The admissible solution must satisfy the EBC: EBC u(O)
Equation
=0
Thus the admissible assumed solution is u(x) = a2x2 + a1x. Weighting functions -} (x, x2) Substitute into the weak form and perform integrations to get: Weight
Equation
x
/
Solving these equations,
-cL2 - 6crL2 - 12Pr 2(r2 + 4r + I)EA o '
-CL2 + 7crL 2 - 12P + 12Pr 4L(r2 + 4r + I)EA o
Substituting into the admissible solution, we get the following solution of the problem:
x(c(2L(6r + 1) - Trx + x)L2 + 12P(2Lr - xr + x)) ux=-'--'--'----'----;;--'-----'------'-'() . 4L(? + 4r + I)EA o .The exact solution of the problem was derived earlier. The linear and quadratic solutions are compared with the exact solution in Figure 2.6. The following numerical values are used: c = 0;
P= 1;
Ao = 1;
L= 1;
E
= 1;
r -- 12
The quadratic solution is reasonably close to the exact solution. Of course the solution can be improved further by starting with a cubic or a higher-order polynomial.
ONE-DIMENSIONAL BVP USINGGALERKIN METHOP
u(x) 1.75 1.5 1.25
1 - - Quadratic
0.75 0.5
- - Exact
0.25
x 0.2
0.4
0.6
0.8
Figure 2.6. Solutions of tapered axially loaded bar subjected to end load
2.3
ONE-DIMENSIONAL BVP USING GALERKIN METHOD
So far only the axial deformation problem has been used to illustrate the Galerkin method. The method in fact is applicable to any differential equation. It is particularly well suited to boundary value problems (BVPs) in which a solution must satisfy differential equations and several boundary conditions over the domain. This section first summarizes the overall solution procedure and then presents several examples of finding approximate solutions of one-dimensional boundary value problems.
2.3.1
Overall Solution Procedure Using Galerldn Method
Given a boundary value problem, the overall procedure for obtaining an approximate solution using the Galerkin method is summarized here. Several examples in this section further clarify the details. (i) Construct the wealcform. 1. Move all terms in the differential equation to the left-hand side. With an assumed solution the left-hand side now represents the residual e(x). 2. Define the total weighted residual. It consists of integral of the left-hand side of the differential equation multiplied by a weighting function Wi' 3. Use integration by parts to incorporate boundary conditions. For a second-order problem, all terms involving second-order derivatives are integrated by parts once. The highest order of remaining terms in the residual should therefore be 1. For a fourth-order problem integration by parts is used twice on all terms involving fourth-order derivatives and once' on the terms involving third-order derivatives. The highest order of remaining terms in the residual should therefore be 2. 4. Identify essential and natural boundary conditions.For second-order problems boundary conditions involving first-order derivatives are natural because they
115
116
MATHEMATICAL FOUNDATION OF THE FINITEELEMENTMETHOD
can be incorporated into the weak form. For fourth-order problems boundary conditions involving second- and third-order derivatives are natural. All other boundary conditions are essential. Incorporate natural boundary conditions into the weak form. (ii) Construct an admissible assumed solution. 1. Start with an assumed solution with some unknown parameters and enough terms in it so that it is possible to evaluate the residual. Any suitable function can be used for the assumed solution. However, since polynomials are easy to differentiate and integrate, it is convenient to start with a polynomial of desired order. 2. Set some of the parameter values such that the essential boundary conditions are satisfied regardless of the values of the remaining unknown parameters. If there are no essential boundary conditions, then the solution already is an admissible solution and we can go to the next step. Otherwise set up equations to satisfy essential boundary conditions. Solve these equations for some of the parameters. Substitute these parameter values back into the starting solution to get an admissible assumed solution. (iii) Set up equations and solve for the unlmown parameters. 1. Determine a set of weighting functions by differentiating the admissible solution with respect to the unknown parameters. 2. Set up a system of equations by substituting each weighting function in tum into the weak form. 3. Solve the system of equations for the unlmown parameters. (iv) Approximate solution. 1. Substitute computed values of the parameters into the admissible solution to get the approximate solution, 2. Check the quality of the solution by substituting it into the differential equation and the natural boundary conditions to see how well it satisfies them.
For convenience the prime and superscript notations for derivatives will be used frequently in the following examples. Thus the first and second derivatives of a function u(x) will be denoted as u'(x) == du(x)/dx and ul/(x) == d 2u/dr. When third- or higher order derivatives are written, the prime notation becomes cumbersome and therefore a superscript notation will be adopted. For example, the third and fourth derivatives of a function u(x) will be denoted as u(3)(x) == d3u/d~ and u(4)(x) == d 4u/dx4.
Example 2.1 Second-Order Equation Find an approximate solution of the following boundary value problem: rul/
+ 2xu' +x = 1;
u(l) = 2;
1
u' (2) + 2u(2) = 5
ONE-DIMENSIONAL BVP USING GALER KIN METHOD
The weak form is derived first followed by detailed calculations of quadratic and cubic approximate solutions. With u(x) as an assumed solution, the residual is e(x)
= u"(x)x'l + 2u' (x)x + x-I
Multiplying by w;(x) and writing integral over the given limits, the Galerkin weighted residual is
Using integration by parts, the order of derivative in x'lwju" can be reduced to 1 as follows:
Combining all terms, the weighted residual now is as follows:
Consider the boundary terms 4wP)u'(2) - wj(l)u'(l)
Given the NBC for the problem: 2u(2) + u'(2) - 5
=0
Rearranging:
(u'(2)
-7
5 - 2u(2))
Given the EBC for the problem: u(l) - 2 = 0 Therefore with admissible solutions (those satisfying the EBC): (wj(l) Thus the boundary terms in the weak form reduce to
4(5 - 2u(2))w i(2) and assuming admissible solutions, the final weak form is as follows:
1 2
4(5 - 2u(2))wi(2) + EBC:
u(l) - 2 = 0
«x - l)w i - x'lu'w;}dx
=0
-7
0)
117
118
MATHEMATICAL FOUNDATION OF THE FINITEELEMENTMETHOD
Quadratic Solution
= a2x2 + alx + ao
Starting assumed solution: u(x)
The admissible solution must satisfy the EBC: Equation
EBC u(l) - 2
=0
Solving this equation, a o = -al - a 2 + 2 Thus the admissible assumed solution is u(x) Weighting functions .... [x - 1, x 2 - l}
= a2x2 + alx -
al
- a 2 + 2.
Substitute into the weak form and perform integrations to get: Weight
Equation
x-I
-~
j~, + 4(5 -
x2 - 1
-~ - l2~a, + 12(5 - 2(a l + 3a 2 + 2)) +
-
2(a[
+ 3a 2 + 2)) + ~ = 0
H= 0
Solving these equations, at = [i6~; a2 = -lstf3 Substituting into the admissible solution, we get the following solution of the problem:
=
u(x)
Cubic Solution
-1090x2 + 4533x + 2329
2886
Starting assumed solution: u(x) = a 3 x3 + a 2x2 + alx + ao
The adrriissible solution must satisfy the EBC: ,/
EBC
Equation
Solving this equation, a o = -a j - a 2 - a 3 + 2 Thus the admissible assumed solution is u(x) = a 3 x3 + a 2x2 + ajx - a l Weighting functions .... {x - 1, x2 - 1, x3 - I}
-
a2 - a3
Substitute into the weak form and perform integrations to 'g~t: Weight
-1.5 Figure 2.7. Error in satisfying the differential equation
Substituting into the admissible solution, we get the following solution of the problem: u(x)
= 22365i3 -
144070x
2
+ 322764x + 13765
107412 Substituting these approximate solutions into the differential equation, we get the error in satisfying the differential equation. This error is plotted in Figure 2.7. u(x)
+
Quadratic
_ 545.t'! 1443
1511.< 962
Cubic
7455x3 _ 72035i' 35804 53706
+ 2329
_ 1090..' 481
2886
+ 26897x + 8951
13765 107412
Error, e(x) + 1992x _ 1 481
22365x3 _ 72035.. ' 8951 8951
+ 62745x _ 1 8951
To demonstrate convergence, solutions up to fifth order are computed. All solutions and their derivatives are compared in Figure 2.8. The higher order solutions are almost on top of each other, indicating the solution convergence. The first derivatives of solutions show greater discrepancy, but they also converge as more terms are included in the assumed solution .
• MathematicafMATLAB Implementation 2.1 on the Book Web Site: Second-order BV? using the Galerkin method 2.3.2
Higher Order Boundary Value Problems
The example problems considered so far in this chapter have all been second-order differential equations. They are called second order because the highest derivative present in the differential equatiori is 2. For the second-order problems, a boundary condition in which only u is specified is essential, and the one involving first derivative of u is called natural. The assumed solutions are not required to satisfy the natural boundary conditions. They are actually incorporated into the weak form and are satisfied approximately with the approximation getting better as the number of parameters in the assumed solution is increased.
119
120
MATHEMATICAL FOUNDATION OF THE FINITE ELEMENT METHOD ------ U2
u(x) - - U3
2.4 2.3
- - U5
2.2 2.1 x
1.2
1.6
. 1.4
1.8
2
du zdx 1
- - U3
~.
0.8-..'....... --U5
0.6 0.4 0.2
1.2
1.6
1.4
x 1.8
2
Figure 2.8. Comparison of solutions and their first derivatives !
The Galerkin method applies to higher order differential equations as well. For example, consider a fourth-order differential equation of the following form: 4u
d + 1= Q. ' dx 4 or
The superscript (4) over u(x) indicates fourth derivative of u(x) with respect to x. With u(x) an assumed solution, the residual is e(x)
= u(4)CX) + 1
Multiplying by wi(x) and writing the integral over the given limits, the Galerkin weighted residual is
ONE·DIMENSIONAL BVP USINGGALERKIN METHOD
Using integration by parts, the order of derivative in
W iu(4)
can be reduced to 2 as follows:
Combining all terms, the weighted residual now is as follows:
Consider the boundary terms
Each one of these terms gives rise to two possibilities:
wi (xo)u" (x o)
Either -u(3)(xO) is known or wi(xO) = 0 Either u"(xo) is known or wi (xo) = 0
W/X,)U(3) (x,)
Either u(3)(x,) is known or wi(x,)
-Wi (x,)u" (x,)
Either -u"(x,) is known or wi (x,) = 0
-W
i(XO)U(3)(XO)
=0
From these requirements the possible boundary conditions are as follows:
NBC
EBe
1
-u(3)(xO) is given
or
2 3
£I" (xo)
is given u(3)(x,) is given
or
= 0 ===> Must satisfy u(x o) boundary condition wi(xo) 0 ===> Must satisfy u'(xo) boundary condition Of- ...w/x,) = 0 ===> Must satisfy u(x,) boundary condition
4
-u"(x,) is given
or
w/xo)
=
wj(x,)
= 0 ===> Must satisfy u'(x,) boundary condition
Thus for a fourth-order problem boundary conditions involving £I and £I' both are essential while those involving £I" and u(3) are natural. A practical problem governed by a fourthorder differential equation is that of beam bending. As will be seen in Chapter 4, for beams, the bending moment is proportional to £I" while the shear force is proportional to £1(3). Thus specified bending moments and shear forces represent natural boundary conditions and displacements and slopes represent essential boundary conditions. The generalization to further higher order differential equations should now be obvious. For a general boundary value problem in which the highest derivative present is of the order 2p (an even order), we should be able to carry out integration by parts p times. Each integration by parts introduces boundary terms that give rise to essential and natural boundary conditions. For the general case the classification of boundary conditions is as follows:
121
122
MATHEMATICAL FOUNDATION OF THE FINITEELEMENT METHOD
. Those with the order from 0 to p - 1 are the essential boundary conditions . . Those with the order from p to 2p - 1 are the natural boundary conditions. The assumed solutions must satisfy all essential boundary conditions for any value of the unknown parameters. Solutions that satisfy the essential boundary conditions and have the necessary continuity for required derivatives are called admissible solutions.
Example 2.2 Fourth-Order Equation Find an approximate solution of the following fourth-order boundary value problem: U(4)(X)
+ 8u"(x) + 4u(x) = 10;
O
with the boundary conditions u(O) = 0; u'(O) = 1; u(S) = 2; u'(S) = O. The superscript (4) on u indicates its fourth derivative. The weak form is derived first followed by detailed calculations of fourth- and fifth-order solutions. With u(x) as an assumed solution, the residual is e(x) = 4u(x)
+ 8u" (x) + u(4)(x) - 10
Multiplying by w;(x) and writing the integral over the given limits, the Galerkin weighted residual is
Using integration by parts, the order of derivative in w;u(4) can be reduced to 2 as follows:
i i
./ 5
(w;u(4»)dx
= w;(S)u(3)(S) -
w;(0)u(3)(0)
i i
+
5
5
(-w;u(3)) dx
5
(-W;U(3») dx
= w;(O)u"(O) -
w;(S)u"(S)
+
(u"w;') dx
Combining all terms, the weighted residual now is as follows: w;(O)u" (0) - w;(S)u" (S) - w;(O)zP)(O) + w;(S)u(3) (S) +
i
5
(2w;(2u + 4u" - S) + u" w;') dx
Consider the boundary terms
Given the EBC for the problem: u(O)
= 0;
u'(O) -1
= 0;
u(S) - 2
= 0;
u'(S)
=0
=0
ONE-DIMENSIONAL BVP USINGGALERKINMETHOD
Therefore, with admissible solutions (those satisfying EBe) w;(O) -7 0;
Assuming admissible solutions, the final weak form is as follows:
is (2wpu + 4u" - 5) + u" W;/) dx
=0
Starting assumed solution: u(x) = a4x4 + a3 x 3 + azxz + a l x + ao
Fourth-Order Solution
The admissible solution must-satisfy the EBC: Equation
EBC.
=
aO = 0
u(O) 0 u'(O) -1
u(5) u' (5)
=0 2 =0
al -1 = 0 ao + 5a l + 25a z + l25a 3 + 625a4 - 2 a l + 10a z + 75a3 + 500a4 = 0
=0
Solving these equations, ao = 0; a] = 1; a z = 25a4 - ~; a3 Thus the admissible assumed solution is 4
u(x) = a4x
.x3
-
.
10a 4.x3 + 125 + 25a4~ -
=0
= ]~5 -10a4 4xz
25 + x
Weighting function -7 {x4 - 10x3 + 25xz} Substitute into the weak form and perform integrations to get: Equation
Weight
]88750a4 _ 23875 -
63
· th'IS equation, . a So1vmg 4
4Z-
0
573 =_}QZO
Substituting into the admissible solution, we get the following solution of the problem:
MATHEMATICAL FOUNDATION OF THE FINITE ELEMENT METHOD
Solving these equations,
Thus the admissible assumed solution is
Weighting functions
-7 {X
4-
1Ox3 + 25x2,.x5 - 75x3 + 250x2 }
Substitute into the weak form and perform integrations to get: Weight
x4
-
Equation
1Ox3 + 25x2
.x5 - 75x3 + 250x2
188750a. 63 2359375"4 63
+ 2359375a,
_ 23875 -
63
+
0
42-
319843750a, _ 873625 693 126-
0
· these equations, . 120581. 3817 SoIvmg Q 4 - - 883350' Q 5 - 146250 Substituting into the admissible solution, we get the following solution of the problem:
Substituting these approximate solutions into the differential equation, we get the error in satisfying the differential equation. A plot of the error shown in Figure 2.9 indicates that both solutions have significant error. Since the differential equation is quite complicated, we need a fairly high-order polynomial to get reasonable solution. To demonstrate convergence, solutions up to 12th order are computed. These solutions are compared in Figure
e(x)
_._- 4th order 100 80
- - 5th order
60 \. 40
\
\
20\"
x Figure 2.9. Error in satisfying differential equation
Figure 2.10. Comparison of solutions and theirfirstderivatives 2.10. The LOth- and 12th-order solutions are fairly close to each other, indicating that they represent good solutions to the problem. . Error, e(x) Fourth-order
• MathematicafMATLAB Implementation 2.2 on the Book Web Site: Fourth-order BVP using the Galerkin method Example 2.3 Third-Order Equation The even-order equations, such as the second and the fourth order, are common in engineering applications. However, the Galerki~ method can be applied to odd-order equations as well. To demonstrate this, we consider the following third-order boundary value problem: U(3)(X)
+ 2u(x)
= ~;
O
125
126
MATHEMATICAL FOUNDATION OF THE FINITE ELEMENTMETHOD
with the boundary conditions u(O) = 1; u(I) = 2; u'(I) = 3. The superscript (3) on u indicates its third derivative. The weak form is derived first followed by detailed calculations of a quadratic solution in Mathematica. With u(x) as an assumed solution the residual is e(x)
= -:t? +2u(x) + u(3)(x)
Multiplying by wi(x) and writing the integral over the given limits, the Galerkin weighted residual is
1
1
(-w ix2
+ 2uw j + w ju(3))dx = 0
Using integration by parts twice on the W j u(3) term, we have
1 1 1
1 1 1
(Wiu(3))dx
1
(-W;'Ul/) dx
= wi(I)ul/(I) -wj(O)ul/(O) +
=u'(O)w;'(O) -
(-w;'ul/)dx
1
+
u'(I)w;'(I)
(u'w/,)dx
Combining all terms, the weighted residual now is as follows:
1 1
u'(O)w;'(O) - u'(I)w;'(l) - wj(O)ul/(O)
The boundary condition u' (1)
+ wi(I)ul/(l) +
(u'w/, -
(:t? -
2u)w j) dx
=0
= 3 can be directly incorporated into the weak form, giving
.
u'(O)w;(O) - 3w;(l) - wj(O)ul/(O)
.
+ )'vi(l)ul/(l) +
1 1
(u'w;' -
•
(:t? -
2u)w)dx
=0
The other two boundary conditions are considered essential: WeakForm[w_, u_] ;= Simplify[«D[u, x] D[w, x])/.x-?O)- «3D[w, x])/.x-? 1) - «D [u , {x , 2}] v) I. x -? 0) +( (D Iu, {x , 2}] w) I. x -? 1) + Integrate [D[u , x] D[w, {x , 2}] - (x-2 - 2u)w, {x , 0, 1}]]
We start with a quadratic polynomial ux with three parameters aD' aI' and a2 as follows: ux = aO + a1 x + a2 x-2; The first task is to make it an admissible solution. We must satisfy the two essential boundary conditions: ebc1 = (uxz . x-« 0) == 1
aO
== 1
ebc2= (ux/vx-» 1) ==2
aO + al + a2 == 2
ONE-DIMENSIONAL BVP USING GALERKIN METHOD
Solving these equations, we get sol = Solve [{ebc1, ebc2} , {aO, a1}J
({aO
~
1, a1
~
1 - a2)}
Substituting the solution into ux gives an admissible solution as follows:
u=ux/. so l I [1J J a2x2 + (l - a2)x + 1 Now we can set up the equation needed to find the remaining coefficient: w=D[u, a2J; eq1=WeakForm[w, uJ ==0
-
1 60 (64 a2 - 147)
.
== 0
Solving this equation, sol = Solve [{eqf.} , {a2}J
{{a2~
l;:n
Substituting the solution into the assumed admissible solution, we have a quadratic approximate solution of the problem as follows: ux = Simplify [ul . sol [[1J J J
1 64(147x2 - 83x + 64) It is possible to obtain an exact analytical solution of the problem as follows: Clear[uJ; soi=DSolve[uJ' [xJ +2u[xJ ==x-2, u Ixl , x] [[1, 1, 2JJ; exactSol = Simplify [soli . Solve [{ (5011. x ~ 0) == 1, (soli .x~ 1) ==2, D[sol, xJI .x~ 1) ==3}J [[1]JJ
127
128
MATHEMATICAL FOUNDATION OF THE FINITEELEMENTMETHOD
u(x)
2
- - - - Approximate
;/
1.8 1.6
- - - Exact
1.4 1.2
/
%' 0.6
0.8
Figure 2.11. Comparison of the approximate and the exact solutions
{D[exactSol, {x, 3}] + 2 exactSol, exactS 011 . x -70, exactSol/ . x -7 1, D[exactSol, xl 1. x -7 1}1ISimplify {x2 , 1,2, 3}
As can be seen from Figure 2.11, the quadratic approximate solution compares very well with the exact solution. A cubic solution will almost be indistinguishable from the exact solution: Needs ["Graphics 'Legend' "] ; __ Plot [{ux, exactSol}, {x , 0, 1}, PlotStyle -7 {Hue [0.6] , Hue [0. 9]}, AxesLabel-7 {"x", "u Cx) "}, PlotLegend -7 {"Approx", "Exact"}, LegendPosition -7 {-1/2, O}, Legendshadow -7 None, PlotRange -7 All] j
2.4
RAYLEIGH-RITZ METHOD
As is the case with the Galerkin method, the Rayleigh-Ritz method also requires an equivalent integral formulation. However, the derivation of this equivalent integral form is based on fundamentally different concepts. For many practical problems,the integral form is based on energy considerations. For structural problems, the potential energy is an example of such an integral form that has been used successfully in developing a large number of finite elements. Similar energy-based integral forms are available for other applications. If the energy function is not known from physical considerations, it may still be possible to derive an equivalent integral formulation using mathematical manipulations. However, such manipulations involve a branch of calculus known as the calculus of variations. A review of basic calculus of variations and mathematical manipulations to derive integral formulations is presented in Appendix B. In the main body of this text the use of the Rayleigh-Ritz method will be restricted to structural problems for which the potential energy function is well known.
RAYLEIGH·RITZ METHOD
"Once the equivalent integral form is known, the steps in the Rayleigh-Ritz method are siillilar to those in the Galerkin 'method. A solution is assumed with some unknown parameters. The assumed solution is made admissible by requiring that it explicitly satisfy the essential boundary conditions. The energy form reduces to a function of unknown parameters once the assumed solution is substiiuted into the energy function. The necessary conditions for the minimum of this function then give the required equations for finding suitable values for the unknown parameters. 2.4.1, Potential Energy for Axial Deformation of Bars Consider a linear elastic bar of any arbitrary cross section subjected to loads in the axial direction only, as shown in Figure 2.1. The area of cross section is denoted by A(x) and could vary over the Iength of the bar. The modulus of elasticity is denoted by E. The bar may be subjected to distributed axial load q(x) along its length. In addition, there may be concentrated axial loads Pi applied at some locations Xi on the bar. The axial displacement 'is denoted by u(x). Denoting the axial stress by 0:, and corresponding strain by Ex, the axial strain energy is defined as follows:
where the integration is over V, the volume of the bar. For axial deformations stress and strain are assumed to be constant over a cross section, and thus the volume integral can be expressed as a one-dimensional integral U=lLx'a:EAdX 2 x xl:.1. Xo
where Xo and x, are end coordinates of the bar. Using Hooke's law and the straindisplacement relationship 0:, = EE., = E du/ dx yields
The potential of the applied loads is equal to the negative of the work done by all external ' applied forces:
x,
W =
L
qu dx + Ipiu(x)
Xo
where u(x) is the axial displacement at the location of the concentrated applied load Pi' Using these expressions, the potential energy for axial deformation of bars is defined as follows: .
II = U - W
(X,
= 1L
Xo
(du)2 (X, : EA(x) dx dx - },. q(x)u(x) dx Xo
I
Piu(x)
129
130
MATHEMATICAL FOUNDATION OF THE FINITEELEMENTMETHOD
Using calculus of variations, it is possible to show that a function u(x) that minimizes the potential energy is a solution of the differential equation governing the axial deformation of bars.
2.4.2
Overall Solution Procedure Using the Rayleigh-Ritz Method
Given a boundary value problem, the overall procedure for obtaining an approximate solution using the Rayleigh-Ritz method is as follows: (i) Obtain the equivalent energy form. For structural problems use the potential energy. For general boundary value problems use calculus of variations to construct an equivalent energy form (see Appendix B). Clearly identify the essential and natural boundary conditions. (ii) Construct an admissible assumed solution. I. Start with an assumed solution with some unknown parameters and enough terms in it so that it is possible to evaluate the terms in the energy form. Any suitable function can be used for the assumed solution. However, since polynomials are easy to differentiate and integrate, it is convenient to start with a polynomial of desired order. 2. Set some of the parameter values such that the essential boundary conditions are satisfied regardless of the values of the remaining unknown parameters. If there are no essential boundary conditions, then the solution already is admissible and we can go to the next step. Otherwise set up equations to satisfy essential boundary conditions. Solve these equations for some of the parameters. Substitute these parameter values ~ack into the starting solution to get an admissible assumed solution. ' (iii) Set up equations and solve for the unknown parameters.
1. Substitute the admissible assumed solution into the energy form and carry out the integration. The resulting expression should be a function only of the unknown parameters. 2. Set up a system of equations by using necessary conditions for the minimum of the energy form. Since the minimum of the energy function corresponds to the solution of the boundary value problem, we get a system of equations by setting to zero the partial derivatives of the energy function with respect to unknown parameters. 3. Solve the system of equations for the unknown parameters. (iv) Approximate solution. I. Substitute computed values of parameters into the admissible solution to get the approximate solution. 2. Check the quality of the solution by substituting it into the differential equation and the natural boundary conditions to see,how well it satisfies them.
RAYLEIGH-RITZ METHOD
2.-4.3
Uniform Bar Subjected to Linearly Varying Axial Load
We now use the Rayleigh-Ritz method to find approximate solutions of a uniform bar (EA constant) fixed at one end and subjected to a static point load at the other end, as shown in Figure 2.3. The bar is also subjected to a linearly varying axial load qEx) = ex, where e is a given constant. The potential energy for the problem is as follows:
r (du)2
I n = 2"EA Jo EEC:
dx
Jor
L
dx - e
xu dx - Pu(L)
u(O) = O.
The simplest possible solution that we can assume is a linear polynomial. Using this as the starting solution, an" approximate solution of the problem is obtained as follows:
Linear Solution
Starting assumed solution: u(x)
= ao + xa 1
The admissible solution must satisfy the EEC: EEC u(O)
Equation
=0
ao
Thus the admissible assumed solution is u(x)
=0
=xa 1•
Substituting the admissible solution into IT and carrying out integration,
For the miriimum of IT, set its derivatives with respect to parameters to 0:
Solving this equation,
Substituting into the admissible solution, we get the following solution of the problem:
u=
(eL2 + 3P)x 3EA
A better solution can be obtained if we start with a quadratic polynomial:
131
132
MATHEMATICAL FOUNDATION OF THE FINITE ELEMENTMETHOD
Quadratic Solution
Starting assumed solution: u(x) == a2x2 + alx + ao
The admissible solution must satisfy the EEC: EEC
Equation
u(O) == 0
Thus the admissible assumed solution is u(x) == a2x2 + alx. Substituting the admissible solution into Il and carrying out integration,
For the minimum of Il, set its derivatives with respect to parameters to 0:
an
Ba,
CL3
-3 + EAa2L2 -
== 0:
an == O:. aa 2
cL4
-4 + 1EAa2L3 -
PL + EAalL == 0
PL2 + EAa lL 2 == 0
Solving these equations,
-7cL2 - 12P 12EA Substituting into the admissible solution, we get the following solution of the problem:
u ==
(l:?? + cL(7L -
3x))x
12EA The exact solution is a cubic polynomial. As demonstrated below, if we start with a cubic polynomial, the Galerkin method finds this exact solution: Cubic Solution
-eL2 - 2P 2EA Substituting into the admissible solution, we get the following solution of the problem:
u>
x(3eL2 - ex2 + 6P) 6EA
Since this is the exact solution, trying any higher order polynomial will not make any difference. Also note that exactly the same solutions were obtained by using the Galerkin method earlier.
2.4.4 Tapered Bar SUbjected to Linearly Varying Axial load Consider a tapered bar fixed at one end and subjected to a static point load at the other end, as shown in Figure 2.4. The bar is also subjected to a linearly varying axial load q(x) = ex, where e is a given constant. The potential energy for the problem is as follows: IT =!E 2 EBC:
rLA(X)(dl~)2dx_e
Jo
u(O)
r
Jo
dx
xudx-Pu(L)
=0
where
A( ) _ A x -
0 -
Ao -AL _ (L + (-1.+ r)x)Ao x --L- -
where Ao is the area of cross section at x Linear Solution
L
= 0, AL is that at x = L, and r =AJA o·
Starting assumed solution: u(x) = ao + xa l
The admissible solution must satisfy the EBC: EBC u(O)
=0
Equation ao
=0
133
134
MATHEMATICAL FOUNDATION OF THE FINITEELEMENTMETHOD
Thus the admissible assumed solution is u(x) = xa l. Substituting the admissible solution into II and carrying out integration,
For the minimum of II, set its derivatives with respect to parameters to 0:
Solving this equation,
{al
-';
} (cL3/3) + PL (l12)LEA o + (l/2)LrEA o
Substituting into the admissible solution, we get the following solution of the problem:
Quadratic Solution
Starting assumed solution: u(x)
=a2x2 + alx + ao
The admissible solution must satisfy the EBC: EBC
Equation
u(O) = 0
ao = 0
Thus the admissible assumed solution is u(x) = a2x2 + alx. Substituting the admissible .solution into II and carrying out integration, I E a2A0 L 3 + 6IE a2A0L 3 + 3' 2 Ea A L2 II -- -4I ca2L 4 - 3I c al L3 + 'ir 2 2 la2 0 + !Ea la 2AoL2 + !rEaiAoL + !EaiAo L - P(a 2L2 +
alP
For the minimum of II, set its derivatives with respect to parameters to 0:
all = 0:
Ba,
all = 0: aa2 Solving these equations, -cL2 - 6crL2 - l2Pr 2(? + 4r + 1)EA o '
-cL2 + 7crL2 - 12P + 12Pr 4L(? + 4r + 1)EA o
COMMENTS ON GALERKINAND RAYLEIGH·RITZ METHODS
Substituting into the admissible solution, we get the following solution of the problem: x(c(2L(6r + 1) - trx + x)L Z + 12P(2Lr - xr + x))
u::::--------;;---------4L(? + 4r + l)EA O
Exactly the same solutions were obtained by using the Galerkin method and are compared with the exact solution in Figure 2.6. ~
MathematicalMATLAB Implementation 2.3 on the Book Web Site:
Tapered bar using the Rayleigh-Ritz method 2.5 COMMENTS ON GALERKIN AND RAYLEIGH-RITZ METHODS 2.5.1 Admissible Assumed Solution It has been mentioned several times already that the assumed solutions must satisfy the essential boundary conditions. This requirement is not difficult to fulfill for one-dimensional problems. As was done in several examples, we simply start with a polynomial of a chosen degree and use the essential boundary conditions to evaluate as many coefficients as possible. Substituting these coefficients into the starting polynomial gives us an admissible assumed solution for the problem. Unfortunately this straightforward procedure does not work for two- and three-dimensional problems. For example, consider a simple situation of a rectangular plate fixed at i.fefi'the end and subjected to loads at the right end, as shown in Figure 2.12. The problem is two dimensional and therefore the displacements are functions of x and y. Similar to the case for one-dimensional problems, we start with an assumed solution for horizontal displacement u(x, y) as a two-dimensional polynomial as follows:
The essential boundary condition is that displacements along the vertical line atx :::: 0 must be zero. To satisfy this requirement, we get the following equation:
u(O, y) :::: ao + azy + asi + .., :::: 0 This is the only equation we have, and thus there is no unique way to find coefficients that will satisfy the equation. One possibility for this example is to set ao :::: az :::: as :::: '" :::: O. This will give us an admissible solution but it may not result in a good approximate solution, because it eliminates a large number of y terms from the assumed solution. . The problem of finding suitable admissible assumed solutions is one of the biggest drawback of the classical methods. This example is'still very simple. Imagine if the support was along an inclined line. How would you find suitable values of coefficients to satisfy the essential boundary condition? As will be seen in the following section, the finite element form of the assumed solution overcomes this difficulty by expressing the solution in terms of nodal unknowns,
135
136
MATHEMATICAL FOUNDATION OF THE FINITEELEMENTMETHOD
y
x
, Figure2.12. Rectangular plate
2.5.2 Solution Convergence-the Completeness Requirement Several examples presented in this chapter have demonstrated that the solutions obtained by using the Galerkin and the Rayleigh-Ritz methods converge to .the exact solution as more terms are added to the' assumed solution. One of the requirements for convergence is that the assumed solutions must be capable of representing all terms in an exact solution. The solution will not converge if a term that is present in the exact solution is not included in the assumed solution. For example, the exact solution for a uniform bar subjected to linearly varying axial load and concentrated tip loading is a cubic function of x. In Section 2.2, the solution of this problem converged to the exact solution when we used a cubic assumed solution. Any complete polynomial of degree larger than 3 would also give exactly the same solution. However, if we start with an assumed solution that skips over, say, the cubic term, as more terms are added, the solution may get better but will never converge to the exact solution. Thus one of the fundamental convergence requirements is that polynomials used in the assumed solutions must be complete. It should also be pointed out that, even though we have considered only polynomial assumed solutions in the examples, there is no reason to restrict ourselves to polynomials. We can use any complete set of functions that satisfies essential boundary conditions. Given enough terms and with a complete set offunctions, the approximate solutions should converge to the exact solution. Example 2.4 As an illustration of lack of convergence, when an incomplete polynomial is assumed, consider the following example:
= 0; u(O) = u(l) = 0 u"(x) + X
O
It can easily be verified that the following is an appropriate weak form for the problem:
1 1
(xw; - u' w;') dx = 0
We first obtain a solution using a complete cubic polynomial:
COMMENTS ON GALERKIN AND RAYLEIGH-RITZ METHODS
'Starting assumed solution: u(x) = a3x3 + azxz + ajx + ao . The admissible solution must 'satisfy the EBC: Equation
EBC u(O)
=0
u(l)
=' 0
ao = 0 ao + a1 + az + a3
=0
Solving these equations, lao -7 0, a 1 -7 -az - a3 } Thus the admissible assumed solution is u(x) = a3x3 + azx2 - azx - a3x. Weighting functions -7 {xz ..;x, x3 - x} Substitute into the weak form and perform integrations to get: Weight Z
Equation
x -x
tz(-4a z - 6a3 -1) = 0
x3-x
fo(-15a z-24a3-4) = 0
Solving these equations, we get az = 0; a 3 =- ~ Substituting into the admissible solution, we get the following solution of the problem:
u(x) = ~(x - x3) By substituting it into the differential equation and the boundary conditions, it can easily be verified that this is the exact solution of the problem. Using complete polynomials, any higher order polynomial will give the same solution. However, if we use an incomplete polynomial, the solution will not be exact. For example, consider a fourth-order polynomial with the linear term missing: Starting assumed solution: u(x) = + + The admissible solution must satisfy the EBC:
Equation
EBC u(O)
u(l)
+
=0 =0
ao = 0 ao + az + a3 + a4 = 0
Solving these equations, lao -7 0, az -7 -a 3 - a4} Thus, the admissible assumed solution is u(x) = a4x 4 + a3x3 - a3xz - a4x2· Weighting functions -7 {x3 - x Z, x4 - x Z } Substitute into the weak form and perform integrations to get: Weight .
MATHEMATICAL FOUNDATION OF THE FINITEELEMENTMETHOD
-H;
is
Solving these equations, we get a3 :::: a4 :::: Substituting into the admissible solution, we get the following solution of the problem: u(x) :::: -{gx2(7x2
-
19x + 12)
Clearly this is not the exact solution. For this example, since we know the exact solution, it is obvious that we can obtain an exact solution by starting with a polynomial having linear and cubic terms only. However, in general, we do not know what terms are present in the exact solution. Therefore, we must use complete polynomials for solutions to converge to the exact solution as more terms are included in the polynomial.
2.5.3
Ga,lerkin versus Rayleigh-Ritz
The Rayleigh-Ritz method requires that given differential equations must first be converted to their equivalent energy form. However, it may not always be possible to obtain this equivalent variational form. For example, the third-order problem in Example 2.3 does not have an equivalent energy form. This apparent disadvantage is not of great concern in actual applications. As pointed out earlier, for a large class of engineering problems the functional form is well known. In fact, in the energy methods of structural mechanics, the functional (usually the potential energy) is used as a basis for developing the governing differential equations. The Galerkin method is clearly the more general of the two methods. It can be applied to any boundary value problem. Furthermore, for those problems for which an equivalent variational form does exist, the Galerkin method gives the same solution as the RayleighRitz method. Thus the Galerkin method is the obvious choice for general applications. The variational methods remain popular in practice because of the large body of existirig literature on the energy methods. For stress analysis applications the governing differential equations are fairly complex. However, the equivalent functional is very straightforward to I write. In these situations the Rayleigh-Ritz method is preferred because it can be applied directly. The Galerkin method would require one to first develop the weak form starting from fairly complex governing differential equations.
2.6
FINITE ELEMENT FORM OF ASSUMED SOLUTIONS
Both the Galerkin and the Rayleigh-Ritz methods are powerful tools for finding approximate solutions to boundary value problems. The quality of the approximation depends on the choice of the assumed solution. However, there are no guidelines for an appropriate choice of assumed solutions, especially for two- and three-dimensional problems. Furthermore, to be admissible, these assumed solutions must satisfy essential boundary conditions. As mentioned before, for complicated problems it may be difficult, if not impossible, to come up with appropriate admissible solutions. Furthermore, since the assumed solutions are defined over the entire solution domain, often a large number of terms must be included in order to represent the solution accurately. The finite element method overcomes these difficulties by introducing two fundamental concepts:
FINITE ELEMENTFORMOF ASSUMED SOLUTIONS
, (i) Solution Domain Is Discretized into Elements. The solution domain is divided into several simpler subdomains called elements. Bach element has a simple geometry so that appropriate assumed solutions can easily be written for an element. The only restriction on the element shape is that it should be possible to carry out the required differentiations and integrations of the assumed solutions over each element. Since each element covers only a small portion of the solution domain, a low-degree polynomial can usually be used to describe the solution over an element. The integral in the weak or the functional form can be evaluated separately over each element and added (assembled) together as follows:
i
o
l
(···)dx=
l~
(···)dx+
x,=O
l~ (···)dx+··· X2
(ii) Coefficients in Assumed Solution over an Element Represent Solution and Its Appropriate Derivatives at the Nodes. In the classical methods the unknown coefficients in the assumed solution do not have any physical meaning. They are just mathematical quantities which when substituted into the assumed solution give an approximate solution. In the finite element method the polynomial coefficients are defined in terms of unknown solutions at some selected points over the element. The locations chosen to define the assumed solution are called nodes. Usually element ends and comers are chosen as nodal locations. The solutions at the nodes are called nodal degrees offreedom. The choice of these nodal degrees of freedom is dictated by the order of derivatives in the essential boundary conditions. Since for a second-order problem the essential boundary conditions do not involve any derivatives, the nodal unknowns for these problems are simply the solution variables. For a fourth-order differential equation, at each node we must choose the solution and its first derivative as degrees of freedom because these are the terms in the essential boundary conditions for this problem. With this choice of nodal degrees of freedom it becomes almost trivial to make the assumed solutions admissible. All that one has to do is to set the corresponding nodal value equal-to the value specified by the essential boundary condition.
Nodal degrees of freedom: quantities needed for BBe Second-order problem: u Fourth-order problem: u and ut Sixth-order problem: u, u', and u"
In the following sections, these ideas are illustrated by writing assumed element solutions for second- and fourth-order ordinary differential 'equations. The solution domain for these one-dimensional problems is a line; therefore the elements for these problems are simply line elements.
2.6.1
linear Interpolation Functions for Second-Order Problems
A two-node line element for a second-order problem is shown in Figure 2.13. The element extends from xl to x 2 and has a length I = Xz - xl' The circles represent nodes. For a secondorder problem the nodal degrees of freedom are the unknown solutions at the nodes and are indicated by u j and uz.
139
140
MATHEMATICAL FOUNDATION OF THE FINITE ELEMENT METHOD
Xl X
Figure 2.13. A simple two-node element for second-order problems
Since there are two nodes on the element, a linear solution can be written over the element by starting from a'linear polynomial with two coefficients and then evaluating these coefficients in terms of the nodal unknowns as follows: U(X) = a o + ajx
= x j; at x = x z;
atx
= u j =::} u j = ao .+ ajx j u(xz) = Uz =::} Uz = a o + ajxZ
u(x l )
Solving the two equations for a o and ai' we get U
-u
a l = - z- - -l Xl -X
z
Thus linear solution over the element in terms of nodal degrees of freedom is as follows:
Collecting together terms involving u j and uz' we can write this solution as follows:
This is the finite element form of a linear solution. It clearly is equivalent to the linear polynomial. However, unlike the polynomial coefficients a o and ai' which do not have any physical meaning, the coefficients u l and U z are the solutions at the two nodes of the element. This is the key feature of the finite element form of the assumed solution and has the following three major advantages: (i) In order to make the solution admissible in the polynomial form, we had to use the essential boundary conditions to set up equations and then solve them to find values for one or more parameters. However, in the finite element form, making the solution admissible is trivial. For example, if the essential boundary condition is u(xz) = 5, then all we have to do is to set U z = 5 and we are done.
FINITE ELEMENT FORM OF ASSUMED SOLUTIONS
u
Element 2 Element 1 -------------- x Figure 2.14. A simple two-element model
(ii) The finite element form is suitable for direct assembly of element equations. Assume we -are using two elements to model a solution domain, as shown in Figure 2.14. The node 2 is common between the two elements. We can get a piecewise linear solution for the entire domain just by making sure that at the common node the two elements use the same nodal value. This simple observation makes it possible to perform computations independently for each element and then simply add the contributions from each element. During this so-called assembly process the nodes common between different elements are assigned the sarrie nodal degree of freedom. (iii) In order to add more terms to the assumed solution, we have two options now. We can use higher order polynomials to derive higher order finite element solutions by following exactly the same procedure used for deriving the linear solution. Alternatively, we can simply use a large number oflinear elements. By using a sufficiently large number of simple elements, we can get reasonably accurate solutions to even complicated differential equations. The finite element assumed solutions are usually written using a matrix notation. A matrix notation is not necessary for this simple element. However, for more complicated elements it is almost impossible to write element solutions concisely' without using the matrix notation. The linearfinite element solution can be written as follows:
where
The Nj(x) are known as interpolation or shape functions and u j are the nodal unknowns and N is a 2 x 1 column vector of interpolation functions. The superscript T denotes the transpose of vector N to make it into a row vector for matrix multiplication with the 2 X 1 vector d of nodal degrees of freedom. . Note that the interpolation function N1 is 1 at node 1 and 0 at node 2 while N2 is 1 at node 2 and 0 at node 1. The function N1 therefore defines the influence of u1 on the
141
142
MATHEMATICAL FOUNDATION OF THE FINITEELEMENTMETHOD
assumed solution and N2 that of u2 . Because of this, the interpolation functions are also known as influence functions.
2.6.2
Lagrange Interpolation
An interpolation function is a smooth function that passes through all data points in a given data set. Thus, treating the nodal locations and unknown solutions at these locations as data points, the problem of writing a finite element solution for second-order differential equations boils down to choosing an appropriate interpolation technique. There are a variety of interpolation methods available in the literature. One simple and well known technique is the Lagrange interpolation. If we have n data points {Xi' u;}, i = 1,2,.:., n [where u(x i) == ui], then the Lagrange interpolation formula for passing a polynomial of degree n - 1 through these points can be written as follows:
I!
u(x)
=I
Li(x)U i == (L I
L2
L3
i=1
where Li(x) are the Lagrange interpolation functions given by the following formula:
L( ·x) I
-N -
=.-
n I!
x-x
-
x-xI
x-x
x-X
x-x +
x-xI!
Xi-XI
Xi-x2 '
xi-xi_ l
Xi-Xi+ 1
Xi-XI!
j 1 1 --= - - x - - x2· · · x - - -i_x - - -i x ···x--
I.
Xi-Xl'
1=1
.
~
The symbol ITindicates a product of terms. For writing the ith interpolation function, the product includes all terms except the ith. Thus there are n - 1 terms in the product and each interpolation function is of degree n - 1, where n is the number of data points. Using this formula, a linear interpolation (n = 2) can be written as follows: U(X)
= N 1(x)u 1 + N2(x )u2 x-x
N (x ) = _ _ 2; 1
XI -X2
N2 (x ) =
X-
X2 -Xl
These are the same shape functions derived in the previous section. A quadratic polynomial (n = 3) can be written as follows: U(X) = N1(x)u 1 + N2(x )u2 + N3(x )u3 x-x x-x
N (x) 1
= __2 X _ _3 ; ~-~
~-~
N (x) 2
x-x x-x =__ 1 X __ 3 ; ~-~
~-~
_ X - Xl. X - X2 ( ) ---X-N3X X3 -Xl
X3 -X2
Example 2.5 Write Lagrange interpolation functions to pass a polynomial through four points. Plot the resulting function if the nodal values and coordinates are as follows:
FINITEELEMENTFORM OF ASSUMED SOLUTIONS
Point
Data Value
o 1 2
1 3
5
I
2:
-2:
We have four data points; therefore n = 4. Each Lagrange interpolation function will be a cubic function. Using the Lagrange interpolation formula, the four interpolation function are N _ (x - XI)(X - X3)(X - X4) z - (xz - xI )(x z - X3)(XZ - X4)
N _ (x - xz)(x - x 3)(x - x 4) . I - (xI - xZ)(xI - x 3)(x l - x 4) ,
'Substituting the numerical values of the coordinates, xl we get Nl
= -!(x -
~)(x - 2)(x - 1);
= 0, Xz = 1, x3 = 2, and x4 = ~,
Nz = Hx - ~)(x '- 2)x
N3 = -(x - ~)(x -1)x;
N4
= !sex - 2)(x -
l)x
Multiplying these interpolation functions with the given nodal values, u l = ~, U z = 1, u3 = 3, and Lt4 = :-~, we get the following cubic function that passes through the given data point:
z
U(X) = _27x3 + l77x _ l13x +
10
20
20
~ 2
The interpolated function is plotted in Figure 2.15. The given data points are shown on the graph with large dots. The function clearly passes through all four data points.
2.6.3
Galerldn Weighting Functions in Finite Element Form
As seen in the previous sections, the finite element assumed solution is expressed in the following form:
u(x) = (N,
N,
... ) (
~ J" N'd
The weighting functions in the Galerkin method are the derivatives of the assumed solution with respect to the unknown coefficients. The unknown coefficients in the finite element
143
144
MATHEMATICAL FOUNDATION OF THE FINITEELEMENTMETHOD
u(x)
3 2.5
2 1.5 1
0.5
x: -0.5
1.5
2
Figure 2.15. Interpolated function passing through given data points
fonn.are the nodal degrees of freedom. Therefore
du w·= -.EN I du, I Thus the interpolation functions also play the role of weighting functions when using the Galerkin method. From this definition of weighting functions, it is easy to see that, if a particular nodal value is specified because of an essential boundary condition, then that degree of freedom is not a variable and hence there is no weighting function corresponding to that degree of freedom. As discussed in Chapter 1, the essential boundary conditions are typically ignored at the element level, and thus all interpolation functions for an element are used as weighting functions to obtain the element equations. It is only after assembly that we tum our attention to imposing essential boundary conditions. The equation corresponding to a specified nodal value clearly does not make sense because of the way the weighting functions are defined. Hence the equations corresponding to specified nodal values must be removed from the global equations before solving for the actual nodal unknowns.
2.6.4
Hermite Interpolation for Fourth-Order Problems
Essential boundary conditions for a fourth-order differential equation consist of the solution and its first derivative. To satisfy these essential boundary conditions at the ends of an element, we must choose the solution and its first derivative as nodal degrees of freedom: Thus the simplest two-node line element for a fourth-order problem has a total of four degrees of freedom, as shown in Figure 2.16. The element extends from xl to x 2 and has.a length 1 = x 2 - xl' For simplicity the left end of the element is assumed at X = 0 and the right end is at x =1. The Lagrange interpolation formula cannot be used to interpolate when we have the solution and its first derivative specified at each data point. We can generate two independent interpolations between (Xl' UI), (X 2' u2 ) } and (Xl' UJ), (X 2' u~)} if we like. However, then there will be no connection between the solution and its first derivative. The appropriate method to interpolate between data values and their first derivatives at each point is known
FINITE ELEMENTFORMOF ASSUMED SOLUTIONS
uz,uz
Ul,UI
6--------------<@ Xj
=0
x2=f
x
Figure2.16. A two-node element for fourth-order problems
as Hermite interpolation. For this two-node element this interpolation is derived by using the basic principles. A general Hermite interpolation formula is given later.
Derivation of Hermite Interpolation for a Two-Node Element Since we have four nodal unknowns, we can derive the finite element form of the solution (i.e., the shape functions) by starting from a cubic polynomial with four coefficients and then evaluating these coefficients in terms of nodal unknowns as follows:
U(x) u'(x)
= ao + ajx + az:? + a3x3
=a
j
+ 2xaz + 3:?a3
= a o;
1= a l
atx = 0:
~ u]
atx = I:
~ Uz = ao + la, + zZa z + Pa3 ;
u
The first two equations give two of the coefficients. Solving the other two equations for az and a3 , we get a - -
z-
3u] - 3llz + 21uI + lu~ .. zZ '
-2u j + 2uz -luI-Iu~
13
Thus a cubic solution overthe element in terms of nodal degrees offreedom is as follows:
u( x)
=u
,
j
+ xU 1 -
X
3 (-2u]
+ 2uz 3
1
'
hI]
I -Iuz)
xZ(3u] - 3uz + 21uI + lu~) zZ
Collecting terms involving the degrees of freedom together, we can write this solution as follows:
145
146
MATHEMATICAL FOUNDATION OF THE FINITEELEMENTMETHOD
The four interpolation functions are
General Formula for Hermite Interpolation Given function values ui and their first derivatives ui at 12 data points xI' x 2" •. , x"' a polynomial of degree 212 - 1 that fits all data can be written as follows: UI I
" u(x) = Ipi(x)U i=1
i
" +I
UI QJx)u; == (PI
QI
... P"
Q,,)
= NTd
i=1
where Pi(x) and Qi(x) are polynomials of degree 212 - 1. These polynomials are expressed in terms of Lagrange interpolation functions Li(x) as follows: P;(X) = (1 - 2L;(x)(x - xi))Lf(x) Qi(x) = (x - x)Lf(x)
Recall that the Lagrange interpolation functions are as follows:
The complete vector of interpolation functions is denoted by N and the corresponding vector of nodal values by d. The functions Pi interpolate between the data values and Qi between the derivative values. Each interpolation function is of degree 212 - 1, where 12 is the number of data points. Note that functions Pi(x) are 1 at the ith data point and zero at . all other data points. The first derivatives of Pi' i = 1, ... , are zero at all data points. On the other hand, the functions Qi(x) are zero at all data points while their first derivatives are 1 at ith point and zero everywhere else. That is, P;(xj ) = oij;
P[(x j )
=0
Qi(x j ) = 0;
Qi(x j )
= oij
where oij is the so-called dirac delta function, which is equal to 1 when i
ootherwise.
= j and equal to
Example 2.6 Using Hermite interpolation, obtain a function that passes through the following two points with the slopes at these points as given in the following table:
FINITE ELEMENT FORM OF ASSUMED SOLUTIONS
Point,
Derivative Value,
Data Value,
o
1
o
1
2
-1
u;
Data point locations: (0, 1I Lagrange interpolation functions for these points:
(L 1 = 1 - x, L z = x] Derivatives of Lagrange interpolation functions:
Derivatives at the given points:
Using these, the Hermite interpolations for data values are
Those for derivative values are
Vector of Hermite interpolation functions
These are the same interpolation functions derived earlier with 1 = 1. The vector of given nodal data values and theirderivatives is
Thus the function that interpolates between these values-is as follows:
u(x)
= (z.x3-3x z+ 1
x 3 - 2x2+x
3x 2 -
2x3 x3-X2 ) [ .
~/]=-3~~+4X2+1
-1
The interpolated function is plotted in Figure 2.17. The given data points are shown on the graph with large dots. The function clearly passes through both data points and has the specified slope of 0 at x = 0 and -1 at x = 1..
147
148
MATHEMATIQAL FOUNDATION OF THE FINITE ELEMENT METHOD
u(x)
2
-1
1.5
0.5
0.2,
0.4
0.6
0.8
x
1.2
Figure 2.17. Interpolated function
Example2.7 Using Hermite interpolation, obtain a function for the following set of data: Point,
Data Value, !Ii
Xi
Derivative Value,
u;
o 1 0 1 2 0 3 0 1 The Hermite interpolation functions to pass a polynomial through three data points and their first derivatives located at x = 0, 1, 3 are as follows: Data point locations: (0, 1, 3} Lagrange interpolation functions for these points: (L J
The vector of function values and their derivatives is ( I . u(x) = N l
19x' lOS
173x4 lOS
0
2
505x 3 lOS
0
0
I). Thus we have
17x 2 4
+ 2N3 + N6 = - - - + - - - - - + - - + 1
The interpolated function is plotted in Figure 2.1S. The given data points are shown on the graph with large dots. The function clearly passes through all three data points and has the specified slope at these points. . u(x)
2 1.5
0.5
x 0.5
1.5
2
-0.5 -1 Figure 2.18. Interpolated function
3.5
149
150
MATHEMATICAL FOUNDATION OF THE FINITEELEMENTMETHOD
2.7
FINITE ELEMENT SOLUTION OF AXIAL DEFORMATION PROBLEMS
As mentioned earlier, using the assumed solution in the form of interpolation functions, it is possible to perform computations independently for each element and then simply add the contributions from different elements. Thus we can employ the standard six-step finite element procedure to obtain approximate solutions:
1. 2. 3. 4. 5. 6.
Development of element equations Discretization of solution domain into a finite element mesh Assembly of element equations Introduction of boundary conditions Solution for nodal unknowns Computation of solution and related quantities over each element
Except for the derivation of element equations, all steps have been discussed in detail in Chapter 1. In this section the procedure for deriving element equations is illustrated considering a two-node element for axial deformation problem. Several numerical examples are then presented that use these equations to solve a variety of axial deformation problems.
2.7.1
Two-Node Uniform Bar Element for Axial Deformations
The simplest element for axial deformations of bars is a two-node line element, as shown in Figure 2.19. The element extends from xl to x 2 and has a length L := x 2 - xl' The circles represent nodes. The unknown solutions at the nodes are indicated by u l and u2 • In addition to the distributed load q(X), the element may be subjected to concentrated axial loads Pi applied at the ends of the element. With these assumptions the governing differential equation over an element is ,I
~ (AE ) +q o., dX dX dU
:=
Any possible concentrated loads at element ends form natural boundary conditions. With the sign convention explained earlier we have _AEdu(x l ) _ . dx -PI'
AE du(x 2 ) dx
:=
P. 2
q
P2
Pi XI
I..
X2
L
~I X
Figure 2.19. A simple two-node element for axial deformation
'.;'
FINitE ELEMENTSOLUTION OF AXIAL DEFORMATION PROBLEMS
The primary unknown is the axial displacement u. Once the displacement is known, axial strain, stress, and force can be computed from the following relationships: d£l
Ex
= dx;
Linear Assumed Solution nodal unknowns. Thus
The assumed solution is a linear interpolation between the
X - X?
£l(x)=
---
( xl -X
z
We will need u'(x) in the later derivation. Differentiating with respect to x, we g~t £l'(x)
= d£l(x) = (_1__l_)(£l l ) dx
Xl -Xz
Ni =_1_ = Xl -X
_~;
z
£lz
Xz -Xl
Nz= _1_ =
L
Xz -Xl
-
(Nf
N2)(u l) =BTd
£lz
~
L
Element Equations Using Galerkin Method The weak form can be written using standard steps of writing the weighted residual, integration by parts, and incorporating the natural boundary conditions. The weighting functions are the same as the interpolation functions Nj • With u(x) an assumed solution the residual is d
e(x) = q + dx (AE£l')
Multiplying by Nj(x) and writing the integral over the given limits, the Galerkin weighted residual is
-L
X 2 (
x,
d ) dx qN j + -.(AE£l')Nj dx
=0
Using integration by parts, AENi(xz)£l'(xz) -AENj(xl)£l'(x l)
Given the NBC for the problem,
Rearranging,
u't» ) -7 J
p
__ 1 .
AE'
+
r'2 (qNi-AEu'N[)dx =0
Jx,
151
152
MATHEMATICAL FOUNDATION OF THE FINITEELEMENTMETHOD
Thus the boundary terms in the weak form reduce to
Assuming admissible solutions, the final weak form is as follows:
With the two interpolation functions, the two equations for the element are as follows:
Noting the property of the Lagrange interpolation functions that N j (xj ) = 1, N, (x 2 ) = 0, N2 (x j ) = 0, and N2 (X 2 ) = I and writing the two equations together using matrix notation, we have
Substituting for u'(x), we have
Jx,r"2 ((NN~ )q - (N' N~) AE(~1 or
where rp is the 2 x 1 vector of element nodal loads (Pj , P2 l and 0 is a 2 x 1 vector of zeros. Rearranging terms and taking nodal degrees of freedom out of the integral sign because they are not functions of x, we have
where
FINITE ELEMENT SOLUTION OF AXIAL DEFORMATION PROBLEMS
Note the careful arrangement of terms when the two equations were written together. Since AEu'(x) is a scalar, the second term inside the integral of the weak form can be written in two different ways as follows:
JXI('<2 (N' N~ ) AEu'(x)dx
, Jx('<2 AEu'(x) (N') N~ dx
or
l
The first form was usedabove to get the equations in the convenient form presented. With the second form we have the following situation:
r:
(N') dx = Jx("? AEB dB dx
JX -AEu'(x) N~ I
T
-
l
'*
Notice that now d cannot be tal
k
=
X
L
2
BAEB T dx
=
XI
rq
Jx.r'2 Nqdx = XI
[
r 2 AE -L12 dx
Jxl • AEJ.. d Jr'2 L2 X XI
[.r -r
q dX ]
rX2~d JX I L ':I X
-
AE -L12 dX] AE ( 1 =_ AEJ.. d . L -1 L2 X
L'2
Jr
':1
X 2 XI
(1)
= qL 21
We now have the two-node element equations for the axial deformation problem: AE ( 1 L -1
. = AE k -( 1 L -1
-1).
1 '
rq
qL = 'T
(1) . 1 ;
r
P
= (PI) P 2
Element Equations Using Rayleigh-Ritz Method The Galerkin method was used to derive element equations in the previous section. The same equations can also be derived using the Rayleigh-Ritz method. The potential energy function for the element is as follows: .
153
154
MATHEMATICAL FOUNDATION OF THE FINITE ELEMENT METHOD
The square of the first derivative of the assumed solution must be written carefully so that the nodal unknowns can be taken out of the integral sign. In order to achieve this goal we proceed as follows: u'(x)
=BTd
(u'(x))z = (U'(XnTU'(X) = (B Td{B Td = dTBB Td
where we have used the rule for the transpose of a product of two matrices (ABl = B TAT. The strain energy term can now be evaluated as follows:
U=
!
i
X2
AEdr BB Td dx = !dT
XI
("2 AEBBT dx d ;: ~l
!dTkd
where
The work done by the distributed load can be evaluated as follows: Wq
= ("2 qudx= ("2 qNTddx=':;;d;:dTrq JX1
JX I
where
r~ =
("2 qNT dx => r q JX1
=
2
(X Nqdx =
Jx, .
(L~:2-!qdX] = q~ (i) 11 qdx L
,/
The work done by the concentrated loads applied at element ends is Wp
=PIu l - Pzuz = (PI
(UI) ;: r~d ;: d Tr
Pz) Uz
p
The potential energy can now be written as follows:
The necessary conditions for the minimum of the potential energy give (see details below)
an = (!kd _ r - r ) + !kd = 0 ad 2 qp 2 Rearranging terms, we get the same element equations as those obtained by using the Galerkin method: kd
=rq + rp =>
AE L
(1-1 -1)( 1 zI) = 2 11) + (PI) Pz U
U
qL (
FINITEELEMENTSOLUTION OF AXIAL DEFORMATION PROBLEMS
The detailed justification for the way the differentiation of the potential energy with respect to the nodal variables is carried out is as follows:
Using the product rule of differentiation,
~~ =(i O)Uk(::~)-rq-rp)+(UI UZ)Uk(~))=O ~~ =(0 1)Uk(::~)-rq-rp)+(UI UZ)uk(n)=O Combining the two equations together, we have
or
Since k is a symmetric matrix, dtGk)
= 4kd, and we have
Thus, even though d is a vector, in the compact matrix form, the expression for the derivative of II with.respect to d appears as if d is a scalar variable. Based on this observation, in the remainder of the book, after writing the energy function, its derivatives with respect to the nodal variables will be written directly without going through a detailed justification every time.
2.7.2 Numerical Examples Example 2.8 Column in a Multistory Building An interior column in a multistory building is subjected to axial loads from different floors, as shown in Figure 2.20. Determine axial displacements at the story levels. Using these displacements, compute the axial strain and stress distribution in the column. Use the following data: Story height, h = 15 ft Story loads: PI = 50,000 lb, Pz = P3
= 40,000 lb, P4 = 35,000 lb
155
156
MATHEMATICAL FOUNDATION OF THE FINITE ELEMENT METHOD
T h
t t t
4
h
3
h
2 x
h
1
UI
Figure 2.20. Column in a multistory building and a four-element model Modulus of elasticity, E = 29 X 106 Ib/irr' Area of cross section, A = 21.S in2 Since during.the derivation of element equations it was assumed that the concentrated loads can only be applied at the element ends, we must divide the column into elements at the story levels. Thus the simplest possible finite element model is the four-element model shown in Figure 2.20. . The concentrated nodal loads will be added directly to the global equations after assembly. All elements have the same length and other properties. Thus the element equations are as follows:
Use pound-inches. The computed nodal displacements will be in inches and the stresses in pounds per square inch. Nodal locations: (0, isn 360, 540, 720)
ElementI: 6
3.51222 x 10 ( -3.51222 X 106
6
-3.51222 x 10 3.51222 X 106
)(£1 1) = (0) £1 2
0
Element 2: 6
-3.51222 X 10 ) (£1 2 ) = (0) 3.51222 X 106 £13 0
6
6
3.51222 X 10 ( -3.51222 X 106
6
Element 3:
3.51222 X 10 . -3.51222 X 10 ) (£1 3 ) ( -3.51222 X 106 3.51222 X 106 £14
= (0) 0
157
FINITE ELEMENT SOLUTION OF AXIAL DEFORMATION PROBLEMS
.
Element 4: 6
3.51222 X 10 ( -3.51222 X 106
-3.51222 3.51222
X X
6
10 ) 106
(u
4) Us
= (0) .
0
Global equations after assembly and incorporating the NBC (placing concentrated applied loads): 3.51222 X 106 . -3.51222 X 106 7.02444 X 106 -3.51222 x 106 6 -3.51222 X 10 o
o o
o
o
o -3.51222 X 106 7.02444 X 106 -3.51222 X 106
o
o o
o
-3.51222 X 106 7.02444 X 106 -3.51222 X 106
]
J.51222 X 106 3.51222 x 106
~:
III
liS
[.
-50~00'
= -40000.
-40000. -35000.
Essential boundary conditions: dof
Value
ul
0
Since the EBC is 0, we simply remove the first row and column to get the final system of equations as follows: 7.02444 X 1'06 -3.51222 X 106 [
o o
o
-3.51222 X 106 7.02444 X 106 -3.51222 X 106 0
-3.51222 X 106 7.02444 X 106 -3.51222 X 106
o. -3.51222 X 106 3.51222 X 106
Z] ](Ll u3 U4 Us
_ -
(-50000.] -40000. -40000. -35000.
Solution for nodal unknowns: dof
x
Solution
ul Uz
0 180 360 540 720
0 -0.0469788 -0.0797216 -0.101076 -0.111041
u3
Ll4 Us
Once the nodal displacements are known, the displacement over any element is obtained from the linear interpolation functions as follows:
x-xz
u(x)= ( - XI -xz
Element 1: Nodes: {XI
-7
0, X z -7 180)
Interpolation functions: NT
= (1 -
I~O' I~O}
158
MATHEMATICAL FOUNDATION OF THE FINITEELEMENTMETHOD
Nodal values: d T = (0, -0.0469788) Solution: NTd = -0.000260993x Element 2: Nodes: (xl -? 180, x 2 -? 360) Interpolation functions: NT = {2 - I~O' I~O - 1) Nodal values: d T = (-0.0469788, -0.0797216) Solution: NTd = -0.000181904x - 0.014236 Element3: Nodes: (Xl -? 360, x 2 -? 540) Interpolation functions: NT ={3 - I~O' I~O - 2) Nodal values: d T = (-0.0797216, -0.101O76) Solution: NTd = -0.000118633 X - 0.0370136 Element 4: Nodes: (xl -? 540, x 2 -? 720) Interpolation functions: NT = {4 - I~O' I~O - 3) Nodal values: d T = (-0.101076, -0.1110411 Solution: NTd = -0.0000553622x - 0.07118 Solution summary:
1 2 3 4
Range
u(x)
0::; x s 180 180::;x::;360 360::; x::; 540 540::; x::; 720
The negative sign indicates compression. The problem is statically determinate, and thus simple application of the static equilibrium condition indicates that the computed axial forces are exact. .
Example 2.9 Tapered Bar Consider solution of the tapered axially loaded bar shown in Figure 2.21. Use a two-node uniform axial deformation element to model the bar and determine the axial stress and force distribution in the bar. Compute the support reactions
FINITEELEMENT SOLUTION OF AXIAL DEFORMATION PROBLEMS
Figure 2.21. Tapered bar
from the axial force and see whether the overall equilibrium is satisfied. Comment on the quality of the finite element solution. Use the following numerical data: E
=70 GPa;
F
= 20kN;
L= 300mm;
where AI and Az are the areas of cross section at the two ends of the bar. The area of cross section can be expressed as a linear function of x using the Lagrange interpolation formula as follows:
Since the displacements are generally small, numerically it is convenient to use newtonmillimeters. Then the computed nodal displacements are in millimeters and the stresses in megapascals. A four-element model is as shown in Figure 2.22. Since the element equations derived earlier are based on the assumption of a uniform cross section, we must assign an average area to each element in the finite element model. Denoting the area of cross section at the left end of an element by Al and that at the right end by A r , the average area for each element is (AI + A r)l2. The concentrated nodal loads will be added directly to the global equations after assembly. Thus, with 1 the length of an element, the element equations are as follows:
Nodal locations: {O, 150.,300.,450., 600.} Areas at the nodes: {2400., 1950., 1500., 1050., 600.} , Average area for each element: (2175., 1725., 1275., 825.)
1---_-_-_---1 2
L 2
L 2
L 2'
Figure 2.22. Four-element model for the tapered bar
159
160
MATHEMATICAL FOUNDATION OF THE FINITEELEMENTMETHOD
Element 1:
Element 2:
805000. ( -805000.
-805000.) (U2) 805000. u3
:=:
(0) 0
Element 3:
595000. -595000.) (U3) ( -595000. 595000. u4
:=:
(0) 0
Element 4:
385000. ( -385000.
-38500,0.) (U 4) 385000. Us
:=:
(0) 0
Global equations after incorporating the NBC:
-1.015 X 106 ,1.82 X 106 -805000. 0 0
[ 1.015 X 10'6 -1.015 X 10 0 0 0
0 -805000. 1.4 X 106 -595000. 0
0 0 -595000. 980000. -38,5000.
ul
-38Jj 385000.
u2 u3 u4
Us
Essential boundary conditions:
dbi
Value
UI
0 0
Us
Incorporating the EBC, the final system of equations is 6
~805000.
1.82 X 10 (
o
-805000. '1.4 X 106 -595000.
1[
0 ] [U 0 -595000. u~:=: 20000. 980000. u4 O.
Solution for nodal unknowns: dof
x
Solution
ul
0 150. 300. 450. 600.
0 0.0129577 0.0292958 0.0177867 0
u2 u3 u4
,us
1
:=:
0 0 20000 0 0
FINITEELEMENTSOLUTION OF AXIAL DEFORMATION PROBLEMS
'Solution over element 1: Nodes: {xl -7 0,x 2 -7l50.} Interpolation functions: NT
= {I. -
0.00666667x, 0.00666667x}
Nodal values: aT = {O, 0.0129577} Solution: NTa = 0.000086385x Solution over element 2: Nodes: {xl -7 150., x 2 -7 300.} Interpolation functions: NT = {2. - 0.00666667x, 0.00666667x - I.} Nodal values: aT = {0.0129577, 0.0292958} Solution: NTa = 0.00010892t - 0.00338028 Solution over element 3: Nodes: {Xl -7 300.,x2 -7450.} Interpolation functions: NT = {3. - 0.00666667x, 0.00666667x - 2.} Nodal values: aT = {0.0292958, 0.0177867} Solution: NT = 0.0523139 - 0.000076727x
a
Solution over element 4: Nodes: {xl -7 450.,x 2 -7 600.} Interpolation functions: NT = {4. - 0.00666667x, 0.00666667x - 3.} Nodal values: aT = {0.0177867, O} Solution: NTa = 0.0711469 - 0.000118578x Solution summary:
1 2 3 4
Range
Solution
0:::;x:::; 150. 150.:::; x:::; 300.. 300.:::; x :::; 450. 45Q.,:::; x :::; 600.
From these displacements we get the following axial strains, stresses, and axial forces: du
Ex
1 2 3 4
= dx;
,0;,
=EEx ;
F:;= Ao:,
Range
E
(T
F
0:::; x:::; 150. 150. :::; x:::; 300. 300. :::; x :::; 450. 450. :::; x :::; 600.
0.000086385 0.00010892 -0.000076727 -0.000118578
6.04695 7.62441 -5.370,89 -8.30047
13152:1 13152.1 -6847.89 -6847.89
The axial forces at the ends must balance the support reactions. With the sign convention for axial forces discussed earlier, the reaction at the left support is the negative of the axial
161
162
MATHEMATICAL FOUNDATION OF THE FINITEELEMENT METHOD
-5000 Figure 2.23. Four-element solution for the tapered bar-e-axial force distribution CT
7.5 5
2.5 +-~---+-----x
100 200 3< 0 400 500 600
-2.5 . -5 -7.5
Figure 2.24. Four-element solution for the tapered bar-stress distribution
force at this point and that at the right support is equal to the axial force. Thus from the axial forces we get the support reactions as Reactions
= {-13152.1, -6847.89}
Sum of reactions = -20000. ,I"
The sum of reactions is equal and opposite to the applied load, and therefore the overall equilibrium is satisfied. Plots of the axial stress and axial force are shown in Figures 2.23 and 2.24. The axial force plot looks reasonable. In the stress plot we expect a discontinuity at the middle because of the concentrated applied force. However, the stress at nodes 2 and 4 should be continuous. A large discontinuity in the stress at these locations indicates that the solution is not very accurate. The following solution is obtained using eight equal-length elements: Range 1 2 3 4 5 6 7 8
0::; x::; 75. 75. ::;x s 150. 150. -s x ::; 225. 225. ::; x ::; 300. 300. -s x ::; 375. 375. s x ::; 450. 450. ::; x ::; 525. 525. ::; x ::; 600.
-7.5 Figure 2.25. Eight-element solution for the tapered bar-stress distribution
The stresses are plotted in Figure 2.25. Because of the constant-area assumption over each element, we still have discontinuities in the stress. However, if we take the average of the stresses from the two elements at common nodes, the solution is very close' to the exact . solution: avg=Map[Apply[Plus, #]/2&, {cr [[{1, 2}]] , o: [[{2, 3}]] , cr[[{6, 7}]], cr[[{7, 8}]]}]
• MathematicalMATLAB Implementation 2.4 on the Book Web Site: Solution ofaxial deformation problems PROBLEMS
Exact Solution of Differential Equations 2.1
A uniform bar is subjected to uniform axial load along its length. The bar is fixed at the left end but there is a gap g between the support and the right end before the load is applied, as shown in the Figure 2.26. Assuming that the applied load is large enough to close the gap, write the governing differential equation with appropriate boundary conditions. Use direct integration to determine the exact solution of the
r
Gap=g
I L
--1
Figure 2.26. Axially loaded bar
163
164
MATHEMATICAL FOUNDATION OF THE FINITE ELEMENTMETHOD
problem. Plot the axial force along the bar. Use the following numerical values:
L
2.2
= 800 mm;
g
=0.25 mm;
E
=200 GPa;
A
=200 mm2 ;
q
= 100 Nlmm
Consider a large plane wall, shown in Figure 2.27, with thickness = 0.4 m, surface area = 20 m2 , and thermal conductivity k = 2.3 W/m . DC. The inside of the wall is maintained at a constant temperature of To = 80DC while the outside loses heat by convection to the surrounding air at Too = IYC with a heat transfer coefficient of h = 24 W/m2 • DC. The governing differential equation for the problem is k
d2T
dr
0< x < L
'
=0'
-k dT(L) dx
= 0.4
=h(T(L) -
T ) 00
Use direct integration to determine an expression for temperature distribution through the wall. What is the rate of heat transfer through the wall?
Inside wall Outside air
Temperature = 80 0 e
. Temperature =15°e
Figure 2.27. Heat conduction through a wall /
2.3
Steady-state heat flow through long hollow circular cylinders can be described by the following ordinary differential equation:
:r (leA d~~)) T(r)
= 71;
=0; T(I) = To
+AQ
where r is the radial coordinate, T(r) is the temperature, k is the thermal conductivity, Q is the heat generation per unit area, A = Znrl: is the surface area, L is the length of the cylinder, rj is the inner radius, and ro is the outer radius. The boundary conditions specify the temperature on the inside and outside of the cylinder, respectively. (a) Show that the following represents an exact solution for the problem for the case when Q = 0: In(r/r)
T(r)
= 71 - (71 - To) In(r: /.) r,
PROBLEMS
(b)
Show that the following is an appropriate weak form for obtaining an approximate solution using the Galerkin weighted residual method:
l
ro (
r,
(c)
dT dw , ) -kA dr d: +AQw j dr
= O'
Using the weak form given in (b), find an approximate solution of the problem with a trial solution of the form T(r) = ao + a j r + a2 ? Assume the following numerical values: ri=lin; L
1f =400°F; To = 100°F k = 0.04 Btu/(h· ft"F)·
r o = 4 in;
= 100 in;
Q
= 0;
'Approximate Solutions Using Galerkin Method
2.4 An engineering analysis problem is formulated in terms of the following ordinary differential equation: d 2u du 'I +x- =)£1-1; d~ dx """ £1(0)
= 2;
u(l)
O
= duel) dx
(a) (b)
What is the order of the differential equation? Is the boundary condition at x = 0 a natural or an essential boundary condition? (c) Is the boundary condition at x = 1 a natural or an essential boundary condition? (d) Obtain a suitable wealeform for the problem. (e) Starting with a linear polynomial, obtain an approximate solution of the problem.
2.5
An engineering analysis problem is formulated in terms of the following ordinary differential equation: 2rc2 sin(rcx)
-£I"
+ ilu -
u(O)
= u(l) = 0
Exact solution:
u(x)
= 0;
O
= sin(rcx)'
Starting with a quadratic polynomial, obtain an approximate solution of the problem. Compare your solution with the exact solution.
165
166
MATHEMATICAL FOUNDATION OF THE FINITEELEMENTMETHOD
2.6
An engineering analysis problem is formulated in terms of the following secondorder boundary value problem:
Zxu" + 3u' u'(l) (a) (b)
2.7
= 4;
=1
and
l
u(2) - 2
=0
Derive a suitable weak form for use with the Galerkin method. Clearly indicate how the boundary conditions will be handled. Starting with a solution of the form u(x) = ao + a1x, obtain an approximate solution of the problem.:
Consider the following boundary value problem:
= 0; u'(7f/2) = 2
u" sin(x) + u' cos(x) + u sin (x) u(7f/4)
=1
and
7f/4 < x < 7f/2
(a)- Verify that the differential equation can be written as follows:
;~ [u' sin(x)] + u sin(x) (b)
2.8
0
Using the Galerkin weighted residual method, obtain an approximate solution of the boundary value problem using a trial solution of the form u(x) = ao+ajx.
Consider the following boundary value problem:
= 1; ;' u(O) = 1;
u" + xu' - 2u u'(O) -
0
=2
Construct a suitable weak form. Using this weak form, obtain a quadratic approximate solution of the problem.
2.9
Consider the following boundary value problem:
ul/+u+x=O with the boundary conditions u(O) = 0, u(l) = O. Construct a suitable weak form. Using this weak form, obtain a quadratic approximate solution of the problem.
2.10
Consider the following boundary value problem:
u" +xu'-u
= 1;
u'(O) - u(O) = 1;
O
=1
Construct a suitable weak form. Using this weak form, obtain a quadratic approximate solution of the problem.
PROBLEMS
2.11
An engineering analysis problem is formulated in terms of the following secondorder boundary value problem: u(x) - u'(x) u'(O)
+ u"(x)
= u(O)
= 1;
and
0
=1
Derive a suitable weak form for use with the Galerkin method. Clearly indicate how the boundary conditions will be handled. Starting with a solution of the form u(x) = ao + a1x, obtain an approximate solution of the problem. 2.12
Use the Galerkin method to find a quadratic solution of the axially loaded bar of Problem 2.1. Verify that this is the exact solution of the problem.
2.13
Use the Galerkin method to find an approximate solution of a uniform bar fixed at both ends and subjected to an axial load q(x) = cx2 leN/m, as shown in Figure 2.28. Assume a quadratic polynomial solution. Determine the axial force distribution in the bar. Plot this force distribution. Determine the support reactions from the force distribution and see whether the overall equilibrium is satisfied. Comment on the quality of the solution. Use the following numerical data: E = 70 GPa;
- - - - - - i...
c
= 200;
L
= 600mm;
A
= 600mm2
X
--~
L
Figure 2.28. Axially loaded bar
2.14
Consider the problem of finding the axial displacement of a truncated solid cone of length L hanging under its own weight and subjected to a downward load F at the tip, as shown in Figure 2.29. The diameter at the top is do and changes linearly to dL at the bottom. The problem is described in terms of the following boundary value problem:
:x
(EA(X)
n
~;) + pA(x) = 0;
A(x) = 4d(x)2; u(O)
= 0;
0
d(x) = do - ¥ x
EA dueL) dx
=F
where p is the weight density and E is the modulus of elasticity. Obtain a quadratic approximate solution using the Galerkin method. Compare this solution with the
167
168
MATHEMATICAL FOUNDATION OF THE FINITEELEMENTMETHOD
exact solution (which can be obtained using Mathematica's NDSolve function). Use the following numerical data:
= I in; d L = 1/4 in; L = 100 in; 6 E = 10 lb/irr': p = l lb/irr'
do
F = 10001b;
F Figure 2.29. Hanging bar
2.15
Derive a weak form for the following sixth-order differential equation. Clearly identify the essential and natural boundary conditions. d 6u(x)
d
4u(x)
_ O'
--6-+--4-+ x dx
2.16
dx
!
,
.
X
o < x < x,
The governing differential equation for the torsion of a thin-walled section with warping restraint can be written as follows: El,/f/III
-
Glorf/'
t(x)
= 0;
O::;,x::;,L
where E is the modulus of elasticity, G is the shear modulus, lw is the warping constant, 10 is the torsional constant, t is the thickness of the section, and ¢ is the angle through which the cross section rotates. Derive an equivalent weak form and' indicate appropriate natural and essential boundary conditions. 2.17
A slider bearing with linear profile is shown in Figure 2.30. The two surfaces shown are moving relative to each other at a constant velocity U and are separated by a viscous fluid with viscosity fl. The governing differential equation for determining pressure in the fluid under the bearing can be written as follows:
~ (h 3dP) + 6fl U dh dx
dx
.
p(O) = peL) = Po
dx
= O.'
O
PROBLEMS
:'where p(x) is pressure in the fluid and hex) is the thickness of the fluid film. Derive a suitable weak form. Starting with a cubic polynomial, obtain an approximate solution using the Galerkin method. Use the following numerical data:
L = 8 in;
ho = 0.0001 in;
U
= 12 in/s;
Po
= 14.7 psi
u
pressure, Po
pressure, Po
x=o
x=L I
Figure2.30. Slider bearing with linear profile
Approximate Solutions Using Rayleigh-Ritz Method
2.18
Repeat Problem 2.12 using the Rayleigh-Ritz method.
2.19
Repeat Problem 2.13 using the Rayleigh-Ritz method.
2.20
Repeat Problem 2.14 using the Rayleigh-Ritz method. Note that the potential energy for the problem is as follows: I1p
U
=U -
W
ILL
=-2
0
(du)2 dx;
EA dx
w=
LL
pAu dx + Fu(L)
Lagrange and Hermite Interpolations
2.21
For a viscous fluid the velocities at various depths are measured as follows: Velocity, m/s Depth below the surface, m
3.75
a
4
1
3.75 1.4
3 1.9
1.75 2.3
Use Lagrange interpolation to obtain an expression for fluid velocity as a function of depth. Plot this relationship and verify that the function passes through all data points.
169
70
MATHEMATICAL FOUNDATION OF THE FINITE ELEMENTMETHOD
2.22 The thickness of a concrete dam at various heights, starting from the bottom, is measured as follows: Thickness, m Height, m
2.81 0
1.79 7.9
1.39 13.7
1.07 18.3
0.8 22
Use Lagrange interpolation to obtain a relationship for dam thickness as a function of height. Plot this relationship and verify that the function passes through all data points.
2.23
Use Hermite interpolation to obtain a function for the following set of data: Point,
xi
Function Value,
Derivative Value,
o
0.1
o
2
0
-1.3
u;
Plot the resulting function and show that it passes through the given points and has the desired slopes at these points.
2.24
Use Hermite interpolation to obtain a function for the following set of data: Point,
o 2.5
Xi
Function Value,
1 0
Derivative Value, u;
o 1.5
Plot the resulting function and show that it passes through the given points and has the desired slopes at these points.
2.25 A flat aluminum bar has constant thickness of 10 mm and a variable profile as shown in Figure 2.31. Use Lagrange interpolation to determine an expression for its area of cross section.
I~
6 @ 15 = 90 rnrn
--1
Figure 2.31.
Two-Node Uniform Bar Element for Axial Deformations
2.26 Find the axial force distribution using only two linear axial deformation elements for the axially loaded bar of Problem 2.1.
PROBLEMS
2.27 :' A bar of constant cross section A and modulus of elasticity E is attached at both ends to rigid supports and is loaded axially by force P as shown in Figure 2.32. Find axial displacement and force distribution using only two linear axial deformation elements. Use the following numerical values: a
= 300 mrn;
b = 600 mrn;
E
= 200 GPa;
A
= 200 mrrr';
P
= 10 kN
Figure 2.32. Axially loaded bar
2.28
Find the axial force distribution using five linear axial deformation elements for the axially loaded bar of Problem 2.13. Assume the load for each element is constant and is equal to the average of the loads at the two ends of the element.
2.29
Find the axial force distribution using three linear axial deformation elements for the axially loaded bar of Problem 2.14. Assume the area for each element is constant and is equal to the average of the areas at the two ends of the element.
2.30
Using the Galerlcin method, derive finite element equations for a two-node axial deformation element assuming that the axial load q varies linearly over the element as shown in Figure 2.33. Assume that EA is constant over the element and that there may be concentrated axial loads Pi applied at the ends of the element.
Pz
x Figure 2.33. Bar element with linearly varying axial load
2.31
A typical finite element for an axial deformation problem involving tapered bars is shown in Figure 2.34. The area of cross section A(x) varies linearly from A[ to A r over the length of the element. Similarly, the applied axial load q(x) varies linearly from q[ to qr over the length of the element. Derive equations for a linear (two-node) finite element using the Rayleigh-Ritz method.
171
MATHEMATICAL FOUNDATION OF THE FINITE ELEMENTMETHOD
Ar
AI
qr
I Figure 2.34. Bar element with linearly varying cross section and load
2.32 A fiat aluminum bar has constant thickness of 10 rom and a variable profile as shown in Figure 2.35. The left end of the bar is fixed. A tensile load of 6 leN is applied at the right end. Determine the axial stress distribution in the bar. Assume the area of cross section changes linearly over each element. Perform a solution convergence study as the number of elements is increased. r-::"1 0 GPO\
I~
6@"15=90mm
~I
Figure 2.35.
.i 'I
CHAPTER THREE
ONE-DIMENSIONAL ,BOUNDARY VALUE PROBLEM
A large number of practical problems are governed by the one-dimensional boundary value problem (lD BVP) of the following form:
d (
dU(X»)
dx k(x)~
+ p(x)u(x) + q(x) =0;
where k(x), p(x), and q(x) are given functions of x and u(x) is the solution variable. Since the differential equation is of second order, for a unique .solution, there must be at least two specified boundary conditions, The boundary conditions of the following form may be specified at one or more points: . Essential boundary
condi~ions
(EBC):
u = specified Natural boundary conditions (NBC): du - dx
= au + fJ
s:: au + fJ.
du
if specified at the left end (x
=xo)
if specified at the right end (x
=XL)
where a and fJ are some specified constants and u is the unknown solution at the point where the NBC is specified. Observe that in the natural boundary condition at the left end the negative of the derivative is specified. The same thing was done in the axial deformation problem in Chapter 2 where the choice was based on the sign convention for the applied nodal forces. In the present case the decision to put a negative sign for the derivative is 173
11 174
ONE·DIMENSIONAL BOUNDARY VALUE PROBLEM
arbitrary. However, with this decision, the finite element equations will come out in the standard form with all terms having consistent signs. Furthermore, when applying this element to physical problems, the sign will make sense with the usual sign convention adopted for those problems. A few selected applications that are govemedby the differential equation of this form are presented in the first section. The second section presents a general finite elementformulation for the problem. The remaining sections in this chapter use this finite element. formulation to obtain approximate solutions for a variety of problems.
3.1 3.1.1
SELECTED APPLICATIONS OF 1D BVP Steady-State Heat Conduction
A simple one-dimensional heat conduction situation is encountered when considering heat flow through large panels or walls. Considering a unit slice of the panel, the heat flow problem through the thickness results in a situation shown in Figure 3.1. The governing differential equation is as follows: d (k,ft dT) dx dx + QA = 0;
where kx = thermal conductivity, A = area of cross section perpendicular to the direction of heat flow, and Q = internal heat source per unit volume. This equation is a special case of the general form with k(x) = k,ft;
=
p(x) 0; ,/
q(x).= QA
The boundary conditions of the following form may be specified at one or more points: T
= specified temperature
-kA ~~
= q specified heat flow
The specified heat flow is considered positive when the heat is flowing into the body and negative if it is flowing out of the body. Thus with a heat flow in the positive x direction we No heat flow
. Figure 3.1. One-dimensional heat conduction
SELECTED APPLICATIONS OF 1D BVP
have the following two situations: At the left end: At the right end:
These boundary conditions are equivalent to the general form with a
3.1.2
= 0 and f3 = q/kA.
Heat Flow through Thin Fins
Thin fins are employed to dissipate heat from a hot body into the surrounding fluid (usually air). A typical situation with uniform fins is illustrated in Figure 3.2. The main mechanism of heat transfer is convection from the large surface area provided by the fins. . A single fin is isolated, and assuming temperature is uniform over the width w of the fin, the problem can be described in terms of the following differential equation:
d (kx'1dx dT) - hPT + QA + hPT dx oo
= 0;
where kx = thermal conductivity, h = convection coefficient, P = surface area per unit length over which convection takes place, A = area of cross section, Too = temperature of surrounding fluid, and Q = internal heat source per unit volume. For the situation shown in Figure 3.2, A = wt, and since the convection takes place at the top, bottom, and sides of the fin, P = 2W + 2t, where t is fin thickness. Generally, Q is zero for heat flow through fins. Compared to the general form, k(x)
= k.~;
p(x)
= -hP;
q(x) '::: QA + hPToo
The boundary conditions ofthe following form may be specified at one or more points:
T
= Tb (specified temperature at the base)
Figure 3.2. Thin fins employed for heat dissipation
175
176
ONE-DIMENSIONAL BOUNDARY VALUEPROBLEM
.Convection heat loss at the free end:
dT .o..kA- = -hA(T-T) dx 00
nsr;
dT hA = -T--dx kA kA
=} -
Compared to the general form,
h k
a=-
and
hT k
f3 =-~
When designing fins, a key quantity of interest is the total heat dissipated by a fin. Once the temperature distribution is known, the total heat loss from a fin can be computed by integrating the heat lost due to convection from the surface of the fin. Thus we have XL
Heat loss =
L
hP(T(x) - Too) dx
Xo
3.1.3 Viscous Fluid Flow between Parallel Plates-Lubrication Problem Viscous fluids are used to reduce friction between moving parts. Actual lubrication problems involve quite complex geometries. However, for a simple one-dimensional analysis the problem can be viewed as a viscous fluid flow between two surfaces that are moving relative to each other. As shown in Figure 3.3, we can assume that the bottom surface is fixed and the top surface is moving at a constant velocity U which is equal to the relative velocity between the two surfaces. The two surfaces are usually at different temperatures, and since the viscosity varies with temperature, the fluid viscosity is typically a function of y. The analysis objective is to determine the fluid velocity profile u(y). . The governing differential equation for .determining fluid velocity profile u(y) can be expressed as follows:
dU) -dP -d ( p(y)-dy
dy
dx
=0
y
u
x Figure 3.3. Velocity profile of a viscous fluidflow between two surfaces
SELECTED APPLICATIONS OF 1D BVP
where j1(y) is the fluid viscosity and dP/dx is the pressure drop in the direction of flow. The pressure drop is measured and must be known in order to determine the velocity' profile. Compared to the general form, k = j1;
p=O;
dP q=-dx
The boundary conditions come from the assumption of no slip at the contact between plates and fluid. Thus we have the following boundary conditions: u(O) = 0
and
u(h) = U
Note that both boundary conditions are of the BBC type.
3.1.4 Slider Bearing A slider bearing with linear profile is shown in Figure 3.4. The two surfaces shown are moving relative to each other at a constant velocity U and are separated by a viscous fluid with viscosity j1. The governing differential equation for determining pressure in the fluid under the bearing can be written as follows: O
where p(x) is pressure in the fluid and hex) is the thiclmess of the fluid film. Compared to the general form, k(x)
= h3 ; -
p(x)
= 0;
q(x)
dh
= 6j1U dx
Outside of the bearing the pressure is equal to the atmospheric pressure, Po. Thus the boundary conditions are as follows (both are of the BBC type): p(O) = peL) = Po
u pressure • po
pressure , Po
x=o
: x=L Figure 3.4. Linear slider bearing
177
178
ONE·DIMENSIONAL BOUNDARY VALUE PROBLEM
F
Figure 3.5. Nonuniform axially loaded bar
3.1.5 Axial Deformation of Bars As seen from Chapter 2, axial deformation of bars is also governed by a differential equation that has the same general form as the ID BVP presented in the first section. For example, the governing differential equation for a nonuniform bar shown in Figure 3.5 is as follows:
d(
dU) + q(x) =0;
O
dx EA(x) dx
where A(x) is the area of cross section, E is Young's modulus, and q(x) is applied distributed load in the axial direction. Compared to the general form, k(x)
=EA;
p(x)
= 0;
The boundary conditions are zero displacement at x U(O)
= 0;
!
q(x)
=q
= 0 and applied axial load F at x = L:
EA(L) dueL) dx
=F
Compared to the general form, the natural boundary condition at x
= L implies a: = 0 and
f3 = FlEA. 3.1.6 Elastic Buckling of Long Slender Bars Consider a long slender bar pinned at one end and simply supported at the other and subjected to an axial load P as shown in Figure 3.6. Under the usual small-displacement assumption, an axial load produces axial displacements only. However, because of small imperfections, long slender bars tend to buckle in a transverse deflection mode. Taking this phenomenon into consideration, the governing differential equation for transverse deflection vex) of slender bars subjected to end axial load P is as follows:
2V)
V !!!... (El d2 ) + P(dd.-2 d2 d2
=0
where E is Young's modulus and I is the moment of inertia. This is a fourth-order differential equation in vIt can, however, be converted into a second-order differential equation
SELECTED APPLICATIONS OF 1D BVP
v
x
Figure 3.6. Buckling of a long slender bar
by defining a new variable y differential equation:
= d2v/dX2. In terms of y we have the following second-order .
d2
dX2 (Ely) + Py = 0 If EI is a function of z, then this equation is not in the form of the general boundary value . problem being considered in this chapter. This can be seen more clearly by expanding the first term by using the chain rule of differentiation, which gives
2y) EI(d + 2 d (El) dy + dX2 dx dx
(p + d2(EI))y =0 dX2
The factor 2 in the second term is not present in the general form since, expanding the first term of the general form of boundary value problem, we have
2U)
!!.(kdU).= k(d dx dx dX2
+ dkdu dx dx
If EI is constant, we have
This is in the form of the general boundary value problem. Thus for a uniform bar compared to the general form
p=P;
k=EI;
q=O
It is shown in Chapter 4 that the bending moment is related to the second derivative of the transverse displacementas follows:
Since the moment is zero at a simply supported end, we get the following boundary conditions for the second-order problem: y(O)
= 0;
y(L)
=0
179
180
ONE-DIMENSIONAL BOUNDARY VALUEPROBLEM
If the axial load P is given, then the problem can be solved in a usual manner to get y(x). However, a more interesting application is to find load P that will cause the bar to buclde. To find the buclding load P, the finite element equations are written and assembled in the usualmanner, However, the global equations result in an eigenvalue problem. The eigenvalues are the buckling load values. The corresponding eigenvectors are the buckling modes. The procedure is illustrated through a numerical example in a later section.
3.2
FINITE ELEMENT FORMULATION FOR SECOND-ORDER 1D BVP
Element equations for a general n-node element for the second-order ID BVP are derived in this section. The procedure is exactly the same as that used in Chapter 2 for deriving equations for a two-node element for axial 'deformations. Each element can have up to n nodes, where n ~ 2. The element extends from Xl to xn and has a length L x n - Xl' The circles represent nodes. The unknown solutions at the nodes are indicated by £II> £1 2 , ••• , £I", as shown in Figure 3.7. Thus over an element we must satisfy d (
dU(X))
dx k(x)~
+ p(x)u(x) + q(x) = 0;
Xl
< X < X"
As discussed in Chapters I and 2, any essential boundary conditions will be imposed after assembling all element equations by setting the corresponding nodal degrees of freedom to the specified values. During the derivation of element equations we therefore do not have to worry about the essential boundary conditions. Any specified natural boundary conditions are included in the weak form, and thus during the derivation of element equations we rriust allow for the possibility of an NBC at the ends of an element. Therefore, during derivation of element equations we consider the following conditions at the ends of an element:
where we have used subscripts on the a and f3 terms to indicate the node number at which an NBC is specified. It should be recognized that in actual numerical solutions of problems any specified NBC 'Will affect the first node of the first element and the last node of the lastelement. Most other elements in the model will have no contributions from these NBC terms.
xi L
·1 x
Figure 3.7. An n-node element for second-order BVP
FINITEELEMENT FORMULATION FOR SECOND-ORDER 10 BVP
Assumed Solution interpolation:
The assumed solution can easily be written using the Lagrange
n
u(x)
= I:N;(x)u(x) = (NI
Nz
i=1
where Ni(x) are the Lagrange interpolation functions
We will need u' (x) in the later derivation. Differentiating with respect to x, we get
u'(x)
= (Nl
Nz
...
UI]
N~) L~Z == B T d
[
un
Element Equations Using Galerkin Method The 'weak form can be written using standard steps of writing the weighted residual, integration by parts, and incorporating the natural boundary conditions. The weighting functions are the same as the interpolation . functions Ni • With u(x) an assumed solution, the residual. is e(x)
= q(x) + p(x)it(x) + k' (x)u' (x) + k(x)u" (x) -
Multiplying by Ni(x) and writing the integral over the given limits, the Galerkin weighted residual is
1~>qNi + puN; + k' a'N, + ku"N;)dx = 0 Using integration by parts, the order of derivative in kN;u" can be reduced to 1 as follows:
1
x " (kN;u")dx
= k(x,,)N;(x
ll)u'(x,)
- k(xl)Ni(xl)u'(x1) +
1
x " (-u'(N;!,'
+ k Nf)) dx
1
1
Combining all terms, the weighted residual now is as follows: k(x,.)N;Cxn)u'(x,,)- k(x1)N,·(xl)U'(x 1) + i'\qN; + puN; - ku'N[)dx
=0
I
Specified values of u' at the end nodes of an element can be directly substituted into the boundary terms to give k(x ll )(f3" + a"u(xn))Ni(x,,) - k(x l)( -/31 - a l u(xl))N;(x l) + iX" (qNi + puNi - ku'N!) dx = 0 XI
181
182
ONE-DIMENSIONAL BOUNDARY VALUE PROBLEM
With the n interpolation functions NI' N2 , ... ,Nn , the system of n equations for the element, written in the matrix form, is as follows: .
As noted earlier, the interpolation functions are such that N, = 1 at xi and 0 at all other nodes. Thus in the boundary terms N] (x]) =Nn(x,) = 1 and all other interpolation functions are zero, giving
Rearranging terms and moving all quantities not involving unknown u to the right-hand side, we have .
Substituting the assumed solution u(x) and its first derivative u'(x), we have
FINITE ELEMENTFORMULATION FOR SECOND-ORDER 1D BVP
Each term in this equation is simplified separately as follows: First term:
where
f" kNiNzdx 1:" kNzNzdx 1
...
"'J
Second term:
where
- f" pN1N] dx - f" pNZN] dx XI
kp
=
[
Xl
L.
·· ·
f" pN]NZdx "'J - f" pNzNz dx ...
-
XI
XI
(x" 1.<1·
x"
q(X)
N. (NIl :Z dx = N"
. ., , . ,
qN dx I
r<" qNzd;
JX1
(X" 1.<1
N d
q "
X
183
184
ONE-DIMENSIONAL BOUNDARY VALUE PROBLEM
Second term on the right side:
Thus the element equations are
As mentioned at the beginning of this section, the k a and TfJ terms will be present in the element equations only if a natural boundary condition is specified for an element. Also for most elements only either the first term or the last term will be nonzero because typically the NBC will be specified at only one end of an element. The equations indicate that, if there is a natural boundary condition specified at node 1 of the element, then a term -
During assembly, the natural boundary condition terms are incorporated directly into the global equations as follows:
NBC atdofi: Add to global k: Add to global r:
k(i, i)
:=
k(i, i) -
rei) := rei) + k(x)f3i
This is a good place to reflect on the decision to arbitrarily assign a negative sign to the derivative term specified at the left end. If it was not done, then it can easily be seen from the above derivation that there will be a positive sign associated with the
FINITEELEMENTFORMULATION FOR SECOND-ORDER 10 BVP
ill
I,
L
x Figure 3.8. A two-node linearelementfor second-order BVP
here conforms to the usually accepted physical notions (such as heat flowing into a body is positive and that out of the body is negative). Explicit integration of the terms in the element equations is possible after one has chosen the number of nodes for an element and the functions k(x), p(x), and q(x) are known.
Explicit Equations for a Two-Node Linear Element For a two-node element, shown in Figure 3.8, the interpolation functions are simple linear functions of x. Furthermore, if we assume that k, p, and q are constant over an element, then it is easy to carry out integrations and write explicit formulas for element equations: Element nodes: (xl' XI + L) Interpolation functions, NT
BT
T
- dN _ -dx-
(_1L 1) L
k = (XI+L(kBBT)dx = k
Jx
=( x1+Z-x
I
k k) ( -rLk _-r rk ip
k
p
rTq
= J(x,+L(_ NNT)d = -3 P x_02 x, (
6
= J(x,+L( !::L} qN)d x = {!::L 2' 2 x,
Thus explicit equations for a linear element are as follows:
Explicit Equations for a Three-Node Quadratic Element With reasonable effort, it is possible to obtain explicit expressions for a three-node element; shown in Figure 3.9, with the middle node placed at the center of the element. As was the case for the linear element, it is also assumed that k, p, and q are constant over an element: Element nodes:
{XI'
4(2xl + L), XI + L}
185
186
ONE·DIMENSIONAL BOUNDARY VALUE PROBLEM
Xl
-j
L
X
Figure 3.9. A three-node quadratic element for second-order BVP
Thus explicit equations quadratic element are as follows: / _Jl!£ _.!d!. 3L
16k _ 3L
_Jl!£ _ 3L
15
1.+&.] &.15 3L
30
15
&.
15
U
.= [!:fl.] ?:!::!I. 6
L~ LJ ~
gp, _Jl!£ _ 3L
3L
15
·U
I
2
3
6
3.2.1 Complete Solution Procedure The approximate solution of any problem governed by a differential equation of the type considered in this chapter can be obtained by using the element equations derived in this section by following the usual finite element steps:
1. 2. 3. 4. 5. 6.
Development of element equations Discretization of solution domain into a finite element mesh Assembly of element equations Introduction of boundary conditions Solution for nodal unknowns Computation of solution and related quantities over each element
·FINITE ELEMENTFORMULATION FORSECOND-ORDER 1D BVP
. The simplest element is obviously the two-node linear element. Since the first derivative of the assumed solution is constant over an element, one must use a fairly large number of linear elements to get accurate solutions. With three-node quadratic elements one typically needs fewer elements. If desired, one can develop a four-node cubic element as well that will require even fewer elements for accurate solution. However, each higher order element obviously requires more effort, and thus, in practice, problems are solved almost exclusively using linear or quadratic elements. Because of well-known difficulties associated with the higher order Lagrange interpolation functions (see Chapter 9), it is not recommended to use elements higher than a cubic based on the formulation presented here. Solutions to many practical problems can be obtained by using the explicit element equations which are based on the assumption of constant coefficients k, p, and q over an element. When these coefficients are not constant, there are two choices. The simplest option is to assign average coefficient values to each element and use the explicit equations given. As the number of elements is increased, the discrepancy between the ayeraged values and the actual coefficient values should diminish. Alternatively, one can include the actual coefficients as functions of x and carry out integrations to establish appropriate matrices in the element equations: kp
=
x"
L
(-p(x))NNT dx;
rq
=
Xl
L X"
q(x)N dx
Xl
Several examples in the following sections illustrate the procedure.
Mathematica Functions for Solution of One-Dimensional Boundary Value Problem For Mathematica solution of one-dimensional boundary value problems using a linear or quadratic element, the following functions can be defined and used in the manner shown in Chapter 2 for the axial deformation problem: BVPiDLinElement[k_, p_, q_, [xt., x2_)] := Module[(L = x2 - xi, ke, re}, ke = ((k/L - (L * p)/3, -(k/L) - (L * p)/6), (-(k/L) - (L * p)/6, k/L - (L * p)/3)}; re = ((L * q)/2, (L * q}!2); (ke, r-s] BVPiDLinSolution[(xL, xz.}, [d.l., d2_)] := Module[(L = x2 - xi, u], n = ((xi + L - x)/L, -(xi - x)/Lj; u = Expand[n.(di, d2)]; (xi ~ x ~ x2, ul] BVPiDQuadElement[k_, p_, q..., (xL, x2_, x3_j] := Module[(L = x3 - xi, ke, rej, . ke = (((7 * k)/(3 * L) - (2 * L * p)/i5, (-8 * k)/(3 * L) - (L * p)/i5, k/(3 * L) + (L * p)/30j, ((-8 * k)/(3 * L) - (L * p)/i5, (i6 * k)/(3 * L) - (8 * L * p)/i5, (-8 * k)/(3 * L) - (L * p)(i5j, (k/(3 * L) + (L * p)/30, (-8 * k)/(3 * L) - (L * p)/~5, (7 * k)/(3 * L) - (2 * L * p)/i5)}; re = ((L * q)/6, (2 * L * q)/3, (L * q)/6); [ke, rej
187
188
ONE-DIMENSIONAL BOUNDARY VALUEPROBLEM
BVP1DQuadSolution[{xL, x2_, x3_}, {dL, d2~ d3_}] :== Module[{L == x3 - xl,u}, n == {(2 * xl + L - 2 * x) * (xl + L - x)/L2 , 4 * (xl- x) * (-xl- L + x)/L2 , (2 * xl + L - 2 * x) * (xl - x)/L 2 ) ; u == Expand[n.(dl, d2, d3)]; {xl.::; x .s x3, ull
MATLAB Functions for Solution of One-Dimensional Boundary Value Problem For MATLAB solution of one-dimensional boundary value problems using linear or quadratic elements, the following functions can be defined and used in the manner shown in Chapter 2 for the axial deformation problem: MatlabFiles\Chap3\BVPIDLinElement.m function[ke, reJ == BVPiDLinElement(k, p, q, coord) % Eke,reJ == BVP1DLinElement(k, p, q, coord) % Generates equations for a linear element for lD BVP %k,p,q = parameters defining the BVP % coord = coordinates at the element ends L == coord(2) - coord(l); ke == [k/L - (L * p)/3, -(k/L) - (L * pY6; -(k/L) - (L * p)/6, kIL - (L * p)/3J; re == [(L * q)/2; (L * q)/2J; Mat1abFiles\Chap3\BVPIDQuadEl~ment.m
function(ke, reJ == BVP1DQuadElement(k,p, q, coord)
% [ke, reJ = BVP1DQuadElement(k, p, q, coord) % Generates equations for ~ qua~ratic element % k,p,q = parameters defiriing the BVP % coord = coordinates at the element ends
Example 3.1 Consider heat conduction through a 5-mm-thick base plate of a 900-W household iron. The base area is 150 em? and the thermal conductivity k == 20 W/m' ·C. The inner surface of the base plate is subjected to uniform heat flux generated by resistance heaters inside, as shown in Figure 3.10. During steady state the outer surface temperature
STEADY-STATE HEAT CONDUCTION
Figure 3.10. Baseplate of a household iron
is measured to be 84°C. Determine the temperature through the base plate. What is the temperature at the inside surface? Heat generation in the base plate is zero. The outside surface has an essential boundary condition. The inside surface has a natural boundary condition. The problem is described in terms of the following differential equation and boundary conditions:
O
= 20W/m. DC;
T(L) = 84
and
150 100
A =--2 m2 ; _ kA dT(O)
dx
5 1000
L = -- m
=900
Compared to the general form, we have
3
k(x)=kA
= 10;
NBC atx
= 0:
p(x)
a=O;
=0;
q(x) = 0
f3 = 900 kA
= 3000
The problem is very simple and an exact solution can readily be obtained by direct integration. However, for illustration purposes, a finite element solution is obtained using only one two-node linear element. The finite element model is as shown in Figure 3.11. The element equations are written by substituting numerical values into the explicit linear element equations presented earlier. The complete one-element solution is as follows: Tz 1
2
x=O
x", 0.005
Figure 3.11. One two-node finite element model
189
190
ONE-DIMENSIONAL BOUNDARY VALUEPROBLIi:M
Element nodes: (xl -7 0, Xz -7 0.005)
(-~g -~~: )(~) =( ~) Global equations before boundary conditions are the same as the element equations. Natural boundary conditions: dof
af3
T1
0
dof
k(x)
-k(x) a
k(x) f3
o
900
3000
Global equations after incorporating the NBC:
Essential boundary conditions: dof
Value
Tz
84
Incorporating the EBC, the final system of equations is (60.)(T1)
= (5940.)
Solution for nodal unknowns: dof
X
TI
Tz
Solution 99.
0.005
84
Solution over element 1: Nodes: (Xl -70, Xz -7 0.005) Interpolation functions: NT = (1. - 200. x, 200. x) Nodal values: aT = (99., 84} Solution: T(x) = NTa = 99. - 3000. X A plot of the solution is shown in Figure 3.12. It can easily be verified that the solution satisfies the differential equation and both the boundary conditions. Thus it is an exact solution and there is no need to increase the number of elements. 3.4
STEADY-STATE HEAT CONDUCTION AND CONVECTION
Example 3.2 A 2-m-high and 3-m-wide plate is at a temperature of 100°C. It is to be cooled by using 20-cm-long and 0.3-cm-thick aluminum fins (Tc = 237 W/m . 0C). The fins extend over the entire width of the plate, as shown in Figure 3.13. Along the plate
STEADY-STATE HEATCONDUCTION AND CONVECTION
T
98 96 94
92 90 88 86 0.001
0.002
0.003
0.004
0.005
x
Figure 3.12. Temperature distribution in the base plate
Figure 3.13. One-dimensional heat transfer through a fin
height the clear spacing between the fins is 0.4 em. The ambient air temperature is 2YC and average convection coefficient is 30 W1m2 . DC. Determine the temperature distribution through the fins. What is the rate of heat transfer from the entire finned surface of the plate? Assuming that over the width of a fin the temperature is uniform and that it varies only along its length, the problem can be treated as one dimensional. Heat generation in the fin is zero. Since the convection takes place at the top, bottom; and sides of the fin, P = 2W + 2t, where t is fin thickness. The base has an essential boundary condition. The outside end of the fin surface has a convection boundary condition. The governing differential equation is as follows: d ( leA dT) dx dx - hPT + hPToo
= 2(W + t); T(O) = 100
P
A
and
= 0;
O::::x<.L
=Wt; leA dT(L) dx
= -M(T -
T ) 00
191
192
ONE-DIMENSIONAL BOUNDARY VALUE PROBLEM
Compared to the general form, we have
k(x)
=leA;
NBC atx
= -hP;
q(x) = hPToo
hA a=-leA;
f3 = hAToo leA
p(x)
=L:
Using the given numerical data,
= 0.3/100 m; L =201100 m; k =237; P= 6.006; A =0.009 leA = 2.133; hP = 180.18; hPT = 4504.5 a =-0.126582; f3 =3.16456
W
= 3 m;
t
h
= 30;
Too
= 30
00
(i) Solution Using Four Linear Elements
A finite element solution using four two-node linear elements is presented. The finite element model is as shown in Figure 3.14. For each element the equations are written by substituting numerical values into the explicit linear element equations presented earlier. The complete four-element solution is as follows: Nodal locations: (0,0.05,0.1,0.15,0.2) Element 1:
(T
45.663 ( -41.1585
-41.1585) 1 ) = (112.613) 45.663 T2 112.613
45.663 ( -41.1585
-41.1585) 2 ) = ( 112.613) 45.663 T3 112.613
45.663 ( -41.1585
-41.1585) 3 ) = (112.613) 45.663 T4 112.613
45.663 ( -41.1585
-41.1585) 4 ) 45.663 Ts
Element 2: ,I
(T
Element 3:
(T
Element 4:
(T
2
3
= (112.613) 112.613
4
Ts
12345 x=O x=O.05 x=O.l x=O.15 x=O.2 Figure 3.14. Four-element model for the fin
STEADY·STATE HEAT CONDUCTION AND CONVECTION
Global equations before boundary conditions: 45.663 -41.1585 [
-41.1585 91.326 -41.1585 0 0
o o
o
o o
0 -41.1585 91.326 -41.1585 0
-41.1585 91.326 -41.1585
t; 225.225 ][TI] [112.613] 'T = 225.225
oo o
3
-41.1585 45.663
T4
t;
225.225 112.613
Natural boundary conditions: dof
a
~
k(x)
-k(x)a
k(x)~
Ts
-0.126582
3.16456
2.133
0.27
6.75
Global equations.after incorporating the NBC: 45.663 -41.1585 [
o o '0
-41.1585 91.326 -41.1585 0 0
o
0 -41.1585 91.326 -41.1585 0
oo o
o
-41.1585 91.326 -41.1585
Tz 225.225 ][TI] [112.613] T = 225.225 3
-41.1585 45.933
T4 Ts '
225.225 119.363
Essential boundary conditions: dof
Value
TI
100
Incorporating the EBC, the final system of equations is 91.326 -41.1585 [
-41.1585 91.326 o -41.1585 0-0
o -41.1585 91.326 -41.1585
[T
Z
0 ] T3 ] 0 -41.1585 T4 Ts 45.933
Solution for nodal unknowns:
. Solution over element 1: Nodes: (Xl ~ 0, Xz ~ 0.05) Interpolation functions: NT
dof
x
Solution
TI Tz T3 T4 Ts
0 0.05 0.1 0.15 0.2
100 73.8496 58.3916 50.2425 47.6187
= (1. -
20. x,20. x)
_
-
[4341.08] 225.225 225.225 119.363
193
194
ONE-DIMENSIONAL BOUNDARY VALUEPROBLEM
Nodal values: d T = {100,73.8496} Solution: T(x) =NTd = 100. - 523.009 x Solution over element 2: Nodes: {xl ~ 0.05,x 2 ~ O.l} Interpolation functions: NT = {2. - 20. x, 20. x - 1:} Nodal values: d T = {73.8496, 58.3916} Solution: T(x) =NTd = 89.3075 - 309.16x Solution over element 3: Nodes: {xl ~ 0.1, ~ ~ 0.15} Interpolation functions: NT = {3. - 20. x, 20. x - 2.} Nodal values: d T ={58.3916, 50.2425} Solution: T(x) = NTd = 74.6897 - 162.981 x Solution over element 4: Nodes: {xl ~ 0.15, x2 ~ 0.2} Interpolation functions: NT = {4. - 20. x, 20. x - 3.} Nodal values: d T = {50.2425, 47.6187} Solution: T(x) =NTd = 58.114 - 52.4766 x A plot of the solution is shown in Figure 3.15. The heat loss over a fin is determined by computing heat loss over each element and then summing element contributions: Heat loss
= L:
L
XI+ !
hP(T(x) - Too) dx
Xl
1 2 3 4
Range
Temperature
Heat Loss
0::; x::; 0.05 0.05::; x::; 0.1 0.1::; x s 0.15 0.15::; x s 0.2
To compute the heat loss from the entire plate surface, we need to determine the total number of fins and multiply the heat loss per fin by this number: Nurnb er 0 f fins =
Heat loss through all fins = 286 X 1408.04 = 402700 W (ii) Solution Using Two Quadratic Elements
To see the effect of higher order elements, now consider a finite element solution using only two quadratic elements. The finite element model still looks the same as shown in Figure 3.14. However, the first three nodes define the first element and the next three the second element. For each element the equations are written by substituting numerical values into the explicit quadratic element equations presented earlier. The complete two-element solution is as follows: Nodal locations: 10,0.05,0.1,0.15, 0.2} Element 1:
Element 2:
Global equations before boundary conditions:
52.1724 -55.6788 -55.6788 123.37 6.5094 -55.6788 0 [ o o 0
6.5094 o . o 0 -55.6788 o 104.345 -55.6788 6.5094 -55.6788 -55.6788 123.37 52.1724 6.5094 -55.6788
T, 300.3 ][Tl] [75.075] T = 150.15 3
T4 Ts
Natural boundary conditions: dof
Q!
f3
le(x)
:-le(x) Q!
le(x) f3
-0.126582
3.16456
2.133
0.27
6.75
300.3 75.075
195
196
ONE-DIMENSIONAL BOUNDARY VALUEPROBLEM
Global equations after incorporating the NBC: 52.1724 -55.6788 6.5094
-55.6788 123.37 -55.6788
o o
o o
6.5094 -55.6788 104.345 -55.6788 6.5094
o o
o o -55.6788 123.37 -55.6788
T 300.3 TI] [75.075] 2 6.5094 T3 = 150.15 -55.6788 [ T4 300.3 52.4424 Ts 81.825
Essential boundary conditions: dof
Value
TI
100
.Incorporating the EBC, the final system of equations is 123.37 -55.6788
o o
(
-55.6788 104.345 -55.6788 6.5094
o -55.6788 123.37 -55.6788
0 ][T2] 6.5094 T3 -55.6788 T4 52.4424 Ts
=(5868.18] -500.79 300.3 81.825
Solution for nodal unknowns: dof
x
TI 0 T2 0.05 T3 /0.1 T4 . 0.15 Ts 0.2
Solution 100 74.074 58.7352 50.6042 47.9969
Solution over element 1: Nodes: (xl --70, x 2 --7 0.05, x 3 --7 0.1) Interpolation functions: NT = (200. x2 - 30. X + 1.;40. x - 400. x2, 200. x2 - 10. x) Nodal values: d T = (100,74.074, 58.7352} Solution: T(x) =NTd = 2117.42x2 - 624.39i + 100. Solution over element 2: Nodes: (xl --7 0.1,x 2 --7 0.15,x3':'" 0.2} Interpolation functions:
NT
= (200.2 -70. x + 6., -400. x2 + 120. x -
8.,200.2 - 50. x + 3.}
Nodal values: d T = (58.7352,50.6042, 47.9969} Solution: T(x) = NTd = 1104.762 - 438.809x + 91.5686
STEADY-STATE HEATCONDUCTION AND CONVECTION
T 100
90 80
70
60 0.05
0.1
0.2
0.15
x
Figure 3.16. Temperature distribution in the fin
A plot of the solution is shown in Figure 3.16. The heat loss over a fin is determined by computing heat loss.over each element and then summing element contributions: Heat loss
= L:
r Jx,
1
hP(T(x) - Tc,,) dx
Heat Loss
Temperature
Range
2
X2
0::; x s 0.1 0.1::; x::; 0.2
Total heat loss
~
2
2117.42x 1104.76x 2
-
624.39x + 100. 438.809x + 91.5686
916.009 477.924
1393.93
(iii) Solution Convergence
The solution convergence can be demonstrated by evaluating solutions using increasing numbers of elements. Figure 3. 17"shows solutions obtained using one through five linear (two-node) elements. Figure 3.18 shows solutions obtained using quadratic (three-node) elements. The convergence is much more rapid in the case of quadratic elements, where, except for the one-element solution, all other solutions are practically indistinguishable from each other. T
....... 1 element
100
-
.. 2 elements
80
70
60 50
x 0.05
0.1
0.15
0.2
Figure 3.17. Comparison of solutions using linear elements
Figure 3.18. Comparison of solutions using quadratic elements
• MathematicafMATLAB Implementation 3.1 on the Book Web Site: Four-quadratic-element solution 3.5
VISCOUS FLUID FLOW BETWEEN PARALLEL PLATES
Example3.3 Determine the velocity profile for flow between two fixed plates shown in Figure 3.19. The fluid temperature varies linearly from 80°F at the bottom to 200°F at the top. The fluid viscosity at different temperatures is as follows: Temperature, OF
J1 x 10-6
80
120 160
16 14 11
200
7
The governing differential equation for determining a fluid velocity profile u(y) is as follows:
!!- (J1(y) dU) _ dP = 0; dy
u(O)
dy
=0
dx
and
u(h)
0
=0
y
x Figure 3.19. Velocity profile of a viscous fluid flow between two plates
Figure 3.20. Viscosity as a function of temperature
Use the following numerical data:
h = 0.3ft;
dP dx
= -2 X 1O-5 1b/ ft3 '
In order to proceed with the solution, we need an expression for p(y). However, the given data show viscosity as a function of temperature. Thus we first use curve fitting to get the following equation for viscosity as a function of temperature: tmData p[T]
The plot in Figure 3.20 shows very good agreement between this function and the given data. Assuming the temperature varies linearly from 80°F at the bottom to 200°F at the top, the temperature as a function of yis as follows:
T(y)
= 400y + 80
Substituting this into the viscosity-temperature relationship, we get the following expression for viscosity as a function of height:
p(y) = -0.00005l- 0.000015y ~ 0.000016 A finite element solution using four two-node linear elements is presented. The finite element model is as shown in Figure 3.21. For convenience the model is shown horizontally. Ul
Uz
2
4
Us
12345 y =0 y =0.075 Y=0.15 Y=0.225 Y =0.3 Figure 3.21. Four-element model
199
200
ONE-DIMENSIONAL BOUNDARY VALUEPROBLEM
Since J1 is not constant, we cannot use the explicit linear element equations given earlier. We must either assign average values to the elements or derive a new k matrix by substituting the J1Cy) expression and carrying out integrations. Here we use the latter approach and derive the specific element equations first. Derivation of Element Equations Element nodes: {Y1' Y2} Interpolation functions, NT
:= (
B T:=dNT/dx:=(-1Y'-Y2
1)
kCy)
:=
pcy) :=
J1Cy);
~
Y2-Y' 0; qCy)
:=
~7 J1(y)dy kk
:=
f2(j1Cy)BBT)dy
[
Y,
T _ -
rq
(Y'-Y2)2'
:=
-dP/dx
~7 J1(y)dy
~7 J1(y)dy
~7 J1(y)dy
(Y1-Yz)(Y2 y,)
(Y'-Y2)2
{1
f Yz( dP N) d _ dP Jy,-dx Y - 2dx
]
(Y1-Y2)(Y2-Y,)
cy 1 -Y2') 2dx 1dP cy1 -Y2 )}
The complete element equations are as follows:
J;;7 J1(y)dy (Y1-Y2)2
( J;;7 J1(y)dy (y, Yz)(Y2-y,)
The numerator in each term of the coefficient matrix is the same and can be written as " follows: . ,
i
Y2 J1Cy) dy
:=
0.0000166667yI
+ 7.5 x 1O-6Yi - 0.000016Y1
y,
- 0.0000166667y~ - 7.5 x 1O-6y~ + 0.000016Y2 Using these element equations, the complete four-element solution is as follows: Nodal locations: {O, 0.075, 0.15, 0.225, 0.3} Element 1: 7
The computed velocity profile is plotted in Figure 3.22. To demonstrate convergence, the velocity profiles are computed using two to five linear and quadratic elements. The resulting profiles are plotted in Figures 3.23 and 3.24. As expected, the quadratic element solution converges much more rapidly.
3.6 ELASTIC BUCKLING OF BARS
Example 3.4 Compute buckling load for a column simply supported at both ends as shown in Figure 3.25. The governing differential equation is as follows: £1 The length L = 10 ft and £1
(~~ ) + Py = 0; = 106 lb . in2 .
yeO)
=y(L) = 0
ELASTIC BUCKLING OF BARS
2 elements 3 elements 4 elements 5 elements
0.005
0.015
0.01
Figure 3.23. Computed velocity profile using linear elements Y
2 elements
0.3
3 elements 0.2
4 elements 5 elements
0.1 -+==----~-------
0.005
0.Ql5
0.01
u
0.02
Figure 3.24. Computed velocity profile using quadratic elements
Compared to the general form, k=EI;
p
e
P;
q=O
Since the buckling load P is unknown, the element matrices kk and k p are computed separately. The matrix k p is multiplied by the unknown load P. Each of the two matrices are assembled in the usual maniier to form global matrices. The final global equations are written in the following form:
This is a homogeneous system of equations and can be recognized as a generalized eigenvalue problem. A nonzero solution for the nodal degrees of freedom d is possible only v
Figure 3.25. Buckling of a long slender bar
203
204
ONE·DIMENSIONAL BOUNDARY VALUEPROBLEM
if the coefficient matrix cannot be inverted. Thus a necessary condition for a non-trivial solution of the equations is that the determinant of the coefficient matrix be zero:
det(lck + Pk p) =
°
For an n-degree-of-freedom system, evaluating this determinant gives an equation called the characteristic equation that is a polynomial of degree n in terms of P. The roots of the characteristic equation represent different budding loads (eigenvalues). Substituting each budding load in the global equations, one can compute the corresponding nodal displacements. These represent the budding modes (eigenvectors). Generally one is interested in the lowest buckling load and the corresponding mode shape. (i) Solution Using Four Linear Elements
For a two-node linear element the finite element equations are derived in the previous section. Taking the unknown P outside as a common factor, the element equations are as follows:
A finite element solution using four two-node linear elements is presented. The finite element model is as shown in Figure 3.26. The element equations are written by substituting numerical values into these element equations. The complete four-element solution is as follows: Nodallocations: Element 1:
to. 30.. , 60., 90., 120.) /
(VI) _P ( 10. 5.)(V1)_(0) 5. 10. v2 - °
33333.3 -33333.3) ( -33333.3 33333.3 v2 Element 2:
(v
33333.3 -33333.3) 2 ) ( -33333.3 33333.3 v3
_
p( 10. 5.)(V2) _(0) 5. 10. 0 v3
-
Element 3:
33333.3 ( -33333.3 y1
-33333.3)(V
3)_p(10.
33333.3
v4
1
Y3
Y2 D
x
2
5.)(
5. 10.
3
Y4
4
Y5
a
12345 =0 X =30 X =60 X =90 X = 120 Figure 3.26. Four-element model
V3) _ v4 -
(0)0
ELASTIC BUCKLING OF BARS
· Element 4:
33333.3 ( -33333.3
-33333.3)(V
4)_p(1O.
33333.3
5.
Vs
5.) (vs - (0)0
10,
4) _ V
Essential boundary conditions: dof
Value
vI
0 0
V
s
Assembling and incorporating the EEC, the final system of equations is
The example corresponds to the classical Euler buclding situation. The buclding load using Euler's equation is 2 7f
.
~I = 685.389
L
The four-linear-element finite element solution is clearly not very accurate. To get a better solution, we must either use more linear elements or employ higher order elements.
205
206
ONE·DIMENSIONAL BOUNDARY VALUE PROBLEM
(ii) Solution Using Four Quadratic Elements
For a three-node quadratic element the finite element equations are derived in the previous section. Taking the unknown P outside as a common factor, the element equations are as follows: ll.
3L 8k
[
8k -:rr: 16k
-:rr:
3L
k
8k -:rr:
sz
:rr:k
v
-15
-~ [V~)+P -15 ]
7k
:rr:
[
V3
2L
L
30
Using four quadratic elements we get a solution that is practically the same as that given by Euler's formula: Nodal locations: (0,15.,30.,45.,60.,75.,90.,105., 12O.} Element 1: [ 777778 -88888.9 11111.1
'11l1I')
[4
2. 16. 2.
-If) [0) ~: ~: ~
111I'J
[4
2. 16. 2.
-lfJ ~: ~: [O~J
"111I'J
(4
2. 16. 2.
-1. J(V
"1111')
[4
2. 16. 2.
-lfJ-("]°°
-88888.9 177778. -88888.9
-88888.9 77777.8
-88888.9 177778. -88888.9
-88888.9 77777.8
v2 - P 2. v3 -1.
=
Element 2: [ 77777.8 -88888.9 11111.1
v4 - P 2. Vs -1.
=
,/
Element 3: ( 77777.' -88888.9 11111.1
11
-88888.9 177778. -88888.9
-88888.9 77777. 8
-88888.9 177778. -88888.9
-88888.9 77777.8
v6 - P 2. v7 -1.
S)
~: ~~
(0)
=
~
Element 4: [ 77777.8 -88888.9 11111.1
v8 - P 2. vg -1.
Essential boundary conditions: dof
Value
VI vg
°°
2. 4.
v8 Vg
ELASTIC BUCKLING OF BARS
207
.Assembling and incorporating the EBC, the final system of equations is
Solution of the eigenvalue problem: Characteristic equation: det[kk - Pkp ]
0 11111.1 -88888.9 155556. -88888.9 11111.1 0
0 0 0 -88888.9 177778. -88888.9 0
0 0 0 11111.1 -88888.9 155556. -88888.9
0 0 0 0 0 -88888.9 177778.
0 0 0 0 0 2. 16.
= 0 gives
-2.6112 x 107 p 7 + 3.48046 X IOI2p 6 - 1.67258 X 1017 p 5 + 3.64352 X I0 21p 4 - 3.75246 X I0 25p3 + 1.75502 X I0 29p2 - 3.19639 X IOnp + 1.47981 x 1035 = 0 Computing roots of the characteristic equation we get the eigenvalues
This value compares very well with the buckling load obtained from Euler's formula. ~
MathematicafMATLAB Implementation 3.2 on the Book Web Site:
Solution of buckling problem
208
ONE-DIMENSIONAL BOUNDARY VALUE PROBLEM
3.7
SOLUTION OF SECOND-ORDER 1D BVP
As a final example, we consider solution of a second-order ordinary differential equation that does not necessarily represent a physical system. If the coefficients in the differential equation are constant, then we can use the explicit linear and quadratic element equations given earlier. For variable coefficients, we can either use approximate values of the coefficients by using average values over an element or derive specific element equations by using the actual coefficients in the element equations and then carrying out exact integrations. The latter procedure is used in the following example. Example 3.5
Find the approximate solution of the following boundary value problem:
l
=x2 ;
p(x)
= 2:
NBC at x
= 1; a = 0;
=1 fJ = 1
q(x)
For this example k(x) is not constant; therefore, we cannot use the explicit expressions of the finite element equations derived in Section 3.2. We must carry out integrations with k =x 2 to establish the kk matrix in the element equations. (i) Solution Using Two Linear Elements A finite element solution using two linear elements is presented in detail. The finite element model is as shown in Figure 3.27. TlIe complete solution details are as follows: Derivation of element equations: Element nodes: {xI' x 2 } Interpolation functions, NT = (
BT =
dNT
dx
=(
_1_ xl-x2 p(x)
.:'2;'
I)
x,-x t
= 1;
q(x)
=1
Ul
Uz
e
Q
1
x=l
2
2 x
= 1.5
Figure 3.27. Two-element model
3
x=2.
SOLUTION OF SECOND-ORDER 10 BVP
¥)
k P
= f2(-INN T)dx = ( ~ x.~·x- ~ T 3 XI
{X2;XI, X2;X I}
r~ = J~2 (IN) dx =
The complete element equations are as follows:
Two-element solution: Nodallocatio~s: (1, 1.5,2) Element 1: Element nodes:
3. ( -3.25
(Xl -7
-7
1.5)
-7
= (0.25)
-3.25)(U 1 ) 3. u2
Element 2: Element nodes: (x2
6. ( -6.25
1, x 2
0.25
1.5,x 3
-6.25) (U 2) 6. u3
-7
2)
= (0.25) 0.25
Global equations before boundary conditions:
3. -3.25 -3.25 _ 9. [ -6.25
°
°
-6.25 6.
][Uu l] = [0.25] 0.5 2
0.25
u3
Natural boundary conditions: dof
a
f3
u3
°
1
dof
k(x)
-k(x) a
°
Global equations after incorporating the NBC:
Essential boundary conditions: dof
Value
k(x)f3
4
209
210
ONE-DIMENSIONAL BOUNDARY VALUEPROBLEM
Incorporating the EBC, the final system of equations is
9. ( -6.25
-6.25)(U Z )
u3
6.
= (3.75) 4.25
Solution for nodal unknowns: dof
x
U1
1
1
Uz
1.5 2
3.28452 4.12971
u3
Solution
Solution over element I: Nodes: {XI --7 1, X z --7 1.5} Interpolation functions: NT = {3. - 2. x, 2. X - 2.} Nodal values: d T = {I, 3.28452} Solution: u(x) = NTd= 4.56904x - 3.56904 Solution over element 2: Nodes: {xl --7 1.5, Xz --7 2} Interpolation functions: NT = {4. - 2. x, 2. X - 3.} Nodal values: d T = {3.28452,4.12971} Solution: u(x) = NTd = 1.69038x + 0.748954 Solution summary: Solution
Range.
1 2
l~x~1.5 1.5
~ X ~
2
4.56904x - 3.56904 1.69038x + 0.748954
A plot of the solution is shown in Figure 3.28. u
4
3.5 3 2.5
2 1.5 1.2
1.4
1.6
1.8
Figure 3.28. Two-linear-element solution
2
x
SOLUTION OF SECOND-ORDER 1D BVP
1
2
3
x=l
x = 1.5
x=2.
Figure 3.29. One-quadratic-element model
(ii) Solution Using One Quadratic Element A finite element solution using only one quadratic element is presented in detail. The finite element model is as shown in Figure 3.29. Note that the model consists of only one element and all three nodes belong to the same quadratic element. Derivation of element equations: Element nodes: {Xl' x,;x, , x 3 }
I
Iati
nterpo anon
BT
fu
. NT nctions,
= dNT = (_X,+3X r;X dx (x,-x =x 2;
k = k
p(x)
( (x,+x 3-2<)(x3-X) (x,-x i 3
4(x,-x)(x-x3) (X,-X3)2 4X) _ 3X'+'<3(X,-X 3)2
4(x,+x3-2x) (.<,-x3)'
3)
le(x)
=
(X'(ilBBT)dx
Jx ,
=
26x,2+8x3x,+6x,' 15x,-15x3
26x,2+8x,x,+6.<,' 15x,-15.<,
16(2x,2+x3x, +2x,2) 15(x,-x3)
6.<,2+8x,x,+26x32 15x,-15'<3
6X,2+8x3x,+26x,2 15.<,-15x,
3.<,2+9x,x, +23x,2 15x,-15.<,
3X12_X3X,+3x32
= J(X, ( -INNT) dx = P x, . _ -
(X'(lN)d
1.<,
¥
X,-X 3 15
XI-Xl ~5
x3-x,
xl-X3
( 31J
15 ) X,-X'}
x3 '
:cr.] 30
8(x,-x,) 15
_ {Xl-XI _;1,( _ X 6 ' 3 Xl
~
15
2(x,-x,) 15
6
The complete element. equations are as follows: 21x,2+13'<3x,+x,2 15x,-15x,
27x,2+6x3X,+7x32 15x,-15x,
27x,2+6x3x,+7x," 15x,-15x,
_ 8(3x,2+4x3x,+3x,2) 15(x,-x3)
[ 7X,2_4x3X,+7x32 30x,-30x,
7X,2+6x,x,+27x," 15x,-15x,
One-element solution: Nodal locations: {I, 1.5, 2} Element nodes: (Xl -) 1, x 2
17
T
_QI
[
15 9
TO
67
TO
184 15 127
_127 15 37
-15
-15
9]
T
3X,2_ X,X,+3x," ] 15x,-15.<,
23x,2+9x,x,+3x," 15x,-15x,
15'<3-15.<,
rTq
(x,-x3)-
= 1; q(x) = 1
[
k
(X'+X3-2x)(~,-X)-)
u
[u 2
-)
[ 61]
1
U3
1.5, x 3
) _
-
;1, 3 1
6
-)
2}
211
212
ONE-DIMENSIONAL BOUNDARY VALUEPROBLEM
The global equations are the same as the element equations. Natural boundary conditions: dof
a
U3
0
(3
dof
k(x)
-k(x) a
k(x) (3
o
4
Global equations after incorporating the NBC:
(1 ~H -~7][~ll-(~] U3 25 15 9
TO
15
15 37
127
2
-
3
5"
-15
"6
Essential boundary conditions: dof
Value
Incorporating the EBC, the final system of equations is
_1~7)(U2)_(H) 37 U 49 5"
3
is
Solution for nodal unknowns: dof
x
,I
ul
1
£!t.
1.5
859
u3
2
859
Solution over elements: Nodes: (XI -? 1, x 2 -? 1.5, x3 -? 2) Interpolation functions: ={2 x
NT
Nodal values: d T = Solution: u(x) .
-
{I, 2;;94,
Solution
w:}
2954 3759
2- 7x + 6, -4x2+ 12x -
8,2x 2- 5x + 3}
NT d -- _2580i' + 10640x _ 7201 859 859 859
Figure 3.30 shows a plot of this solution.. (iii) Solution Convergence
To demonstrate convergence, the solutions are computed using one to four linear and quadratic elements. These solutions are plotted in Figures 3.31 and 3.32. As expected, the quadratic element solution converges much more rapidly.
SOLUTION OF SECOND·ORDER 1D BVP
u
4 3.5
3
2.5 2
1.5 1.2
1.4
1.8
1.6
x
2
Figure 3.30. One-quadratic-element solution
u .._-_.,. 1 element
4 - - 2 elements
3.5
- - 3 elements'
3
2.5
- - 4 elements
2
1.5
x 1.2
1.4.
1.6
1.8
2
Figure 3.31. Comparison of solutions with one to four linear elements
u
4.5
~
4
- - 1 element - - 2 elements
3.5 - - 3 elements
3
- - 4 elements
x
1.2
1.4
1.6
1.8
Figure 3.32. Comparison of solutions with one to four quadratic elements
213
214
ONE-DIMENSIONAL BOUNDARY VALUEPROBLEM
3.8
A CLOSER LOOK AT THE INTERELEMENT DERIVATIVE TERMS
A sharp reader probably has noticed that in the finite element solutions presented in this and the previous chapter, from a mathematical point of view, the boundary terms arising from the integration by parts are not handled precisely. This was done deliberately to avoid a potentially confusing discussion that actually does not impact the finite element solution procedure. For completeness this discussion is taken up in this section. Recall from Section 3.2 that, after integration by parts, the weighted residual over a general n-node element is expressed as follows:
If there are specified values of u' at the end nodes of an element, then we have
u'(X,)
=/3n + Q!IIU(X,)
Substituting these into the boundary terms, we obtained the following weak form:
Precisely speaking, this form is valid only if each element has natural boundary conditions specified at both ends. If we look back at all the examples presented in the previous sections, it is not true even for one example. Some examples do not have any NBC terms at all while others have an NBC specified at only one end of one element. Thus the question is how do we justify this treatment of interelement boundary terms? The answer is that, during assembly, when element equations are added together, these interelement terms from adjoining elements cancel each other and we are left with terms only at the ends of the solution domain. If an NBC is specified at an end, then we simply use that in the equations and we are in good shape. If an EBC is specified at the end, then the derivative term remains unlrnown. However, since we do not use the equation that corresponds to the degree of freedom with specified EBC, the term that is left out does not cause any problems. For structural problems this unknown term is in fact the reaction corresponding to a node where displacement is prescribed. This point was discussed using physical arguments in Chapter I where we introduced the procedure for handling essential boundary conditions. Here you see the mathematical reasoning for this treatment. To see this more clearly, consider a two-node linear element with nodes at Xl and x 2 and degrees of freedom ul and u2 . Without introducing the NBC terms in the weighted residual, the two equations for the element are as follows:
A CLOSERLOOK ATTHE INTERELEMENT DERIVATIVE TERMS
-k(XI)UI(XI)) (X2 (NI) (NI) ( k(N;) '()d _ ( k(xZ)u'(XZ) + J<, q NZ + P NZ U x) - N~ U x . x -
(0)°
The integral terms have been evaluated earlier in the explicit form, and thus we have the following element equations:
where L is element length. To see cancellation of the boundary terms at intermediate nodes, we use only two such elements to obtain a solution of the following boundary value problem: ul/(X) + 1 = 0; u'(O) - 2u(0)
O
=1;
u(I) = 1
We clearly have a natural boundary condition specified at x condition at x = 1. Compared to the general form, we have
k = 1;
°
and an essential boundary
q=I
p=O;
NBC atx = 0:
=
- u'(O)
= -2u(0) -1
=}
a
= -2,
fJ =-1
The two-element finite element model is as shown in Figure 3.33. Substituting the numerical values, the finite element equations for the two elements are as follows: Element 1:
Assembling the element equations, we get the following global system of equations:
Uz a
1
3
2
x=l
x=O 1 2
x=Figure 3.33. Two-element model
215
216
ONE-DIMENSIONAL BOUNDARY VALUEPROBLEM
Clearly at node 2 the derivative terms from the two elements cancel each other. From the EBC we know u(l) == u 3 = 1. However, u/(l) is not known. From the NBC at x = 0 we know that -ul(O) = -2u(O) - 1 == -2u l - 1. Thus we have
-z ~21(~:~)=[i]+(-2Ub -1) (:2 o -2 2 1 1 ul(l)
4
These equations involve only two unknown nodal degrees of freedom which can be solved for by considering only the first two equations:
1) -2 0 )(U 2 (-2 4 -2 L~2 =(!)1 + (-2U0 -1) l
This clearly shows that in order to solve for the nodal unknowns we do not need to know the derivative term at the location that corresponds to an essential boundary condition. For actually solving for ul and u2 , we rearrange the two equations as follows:
Moving unknowns to the left side,
Thus the final system of equations, after incorporating all boundary conditions, is as fol/ lows:
(-~ ~2 ) C:~) =(
-l)
This example clearly demonstrates the validity of the procedures used in the finite element solution and should dispel any lingering doubts that a reader might have about the nature of finite element computations. Exactly the same arguments can be made by considering two- and three-dimensional problems. However, in those situations we are obviously dealing with line and surface integrals to evaluate boundary terms, and thus it is not easy to demonstrate these details through simple calculations. PROBLEMS . Second-Order 1D BVP
3.1
An engineering problem is formulated in terms of the following boundary value problem:
1
PROBLEMS
with the boundary conditions u'(l) = 1 and u(3) = 1. Express the problem in the form of the general boundary value problem presented in this chapter. (Hint: By expanding the derivative, you can easily see that (dldx)(~x2u') = ~x2u" + xu'.) Clearly identify k.(x), p(x), and q(x) terms. Classify the boundary. conditions into the essential and the natural types. For the natural boundary conditions identify the ex and,B terms. Selected Appllcationa of 1D BVP
3.2
The door of an industrial gas furnace is to be designed to reduce heat loss to no more than 1200 W/m2 • A preliminary design calls for a 200-mm-thick insulation sandwiched between a 6.25-mm-thick stainless steel sheet on the interior surface and a 9.5-mm sheeton the outside. The thermal conductivity of steel is 25 W/m' K and that of insulation is 0.27 W/m· K. The convection coefficient of the inside surface is 20 W1m2 • K and that of the outside surface is 5 W1m2 . K. The inside temperature of the furnace is at 1200°C while the surroundings are at 20°C (Figure 3.34). Use three linear elements, one through each layer, to determine the temperature through the door. Does the design meet the stated heat loss goal? Use the simplest finite element model possible based on linear elements.
1200"C
20"C
Figure 3.34. Furnacedoor
3.3
Circular pin fins are used to dissipate heat from an electronic device. A typical fin is as shown in Figure 3.35. The length of the fin is 2.5 em and its diameter is 0.25 em. The thermalconductivity of fin material is 396 W/rn- K and the convection coefficient around the circumference and the end is lOW 1m2 • K. The base of the fin is at a temperature of 9SOC and the surrounding air temperature is 2SOC. Determine temperature distribution in the fin. Calculate the heat loss through the fin. (a) Use four linear elements. (b) Use two quadratic elements.
Figure 3.35. Pin fin
217
218
ONE-DIMENSIONAL BOUNDARY VALUEPROBLEM
3.4
A 5-cm-Iong turbine blade is exposed to combustion gases at 900·C (Figure 3.36). The area of cross section of the blade is 4.5 cm2 and its perimeter is 12 em. The blade is made of high-alloy steel with thermal conductivity k = 25 W1m . K. The convection coefficient for the blade surface and the end is 500 W/m2 . K. The temperature of the blade at the attachment point is 500·e. Determine the temperature distribution in the blade. (a) Use three linear elements. (b) Use one quadratic element.
Figure 3.36. Turbine blade
3.5
Consider solution of the tapered axially loaded bar shown in Figure 3.37. Divide the bar into two finite elements and determine the axial force distribution in the bar. Plot this force distribution. Compute the support reactions from the force distribution and
Figure 3.37. Axially loaded bar
PROBLEMS
see whether the overall equilibrium is satisfied. Comment on the quality of the finite element solution. Use the following numerical data: E = 70 GPa;
(a) (b)
F = 20 kN;
L = 300 mm;
2
AI = 2400 rnrn ;
A 2 = 600 mrrr'
Use two linear elements. Use two quadratic elements.
3.6 Consider solution of the axially loaded bar shown in Figure 3.38. The two end segments have linearly varying areas of cross section and the middle segment is of uniform cross section. Divide the bar into four equal-length finite elements and determine the axial stress and force distribution in the bar. Compute the support reactions from the axial forces and check to see if the overall equilibrium is satisfied. Plot the stress distribution and comment on the accuracy of the finite element solution. Use the following numerical data: E = 70 GPa;
L = 300 rnrn;
P = 20 kN;
AI = 2400 mm
/I.ca--- L ~ L ~
2
;
A 2 = 600 mrrr'
2L
Figure 3.38. Axially loaded bar
3.7 Determine the axial stress distribution in a bar that is rotating at 500 rpm as shown in Figure 3.39. The problem can be treated as one dimensional with the governing differential equation as follows:
~ (EA dU) + pAxw2= 0; dx:
u(O)
dx
= 0;
EA dueL) dx
=0
Figure 3.39. Rotating bar
0
L
219
220
ONE·DIMENSIONAL BOUNDA,RY VALUEPROBLEM
where x is the coordinate along the axis of the bar, u(x) is the axial displacement, L = length of the bar, E = Young's modulus, A = area of cross section, p = mass density, and w = angular velocity in rad/s. The axial stress is ~ = E du/dx. An exact analytical solution of the problem is
~,Exact =
pw2
2
2
(L -
2")
Compare solutions with one, two, and three elements. Use the following numerical data: L
3.8
= 80cm;
E
= 200 GPa;
A
= 250mm2 ;
p
= 7850 kg/nr'
A slider bearing with stepped profile is shown in Figure 3.40. The two surfaces shown are moving relative to each other at a constant velocity U and are separated by a viscous fluid with viscosity u. The governing differential equation for determining pressure in the fluid under the bearing can be written as follows:
dh
3d P) .:!-(h + 6jJ.U dx dx dx p(O)
O
= 0;
= peL) = Po
where p(x) is the pressure in the fluid and hex) is the thickness ofthe fluid film. Obtain an approximate solution using two linear elements. Use the following numerical values:
jJ.
= 1O-4 lb · s/ft2 ;
L = 8 in;
ho = 0.0001 in;
U
= 12 in/s;
Po
= 14.7 psi
u
pressure, Po
pressure, Po
x=L
x=o
Figure 3.40. Slider bearing with stepped profile
Solution of Second-Order 10 BVP. 3.9
Use three linear finite elements to find an approximate solution of the following boundary value problem:
!u" - u - 4 =0; with the boundary conditions u/(l)
1
= 1, u(4) = 1.
PROBLEMS
3.io
Find an approximate solution of the following boundary value problem: ?
=0
xt u" + 2xu' - xu + 4
with the boundary conditions u(I)
or
d(x 2u') - - - xu + 4 dx
= 1, u'(3) -
= 0;
l
2u(3) = 2.
. 3.11 Find an approximate solution of the following boundary value problem using linear and quadratic finite elements: _utI + 1f2 U u(O)
-
21f2 sinorx)
= 0;
O
= u(I) = 0
Exact solution: u(x)
= sinorx)
3.12 In spherical coordinates the electromagnetic potential V is expressed as a function of 8 by the following boundary value problem:
:8 (Sin8~~) = 0; V(a)
= Vo;
8 1f12
a< <
V(1fI2) = 0
where a is a given angle less than tt and Vo is the specified potential at this angle. Assuming a = 1f/4 and Vo = 1 and using two linear finite elements, determine the potential V at 8 = 31f/8.
221
CHAPTER FOUR
..
TRUSSES, BEAMS, AND FRAMES
Many structural systems used in practice consist of long slender members of various shapes. Common examples are roof trusses, bridge supports, crane booms, and antenna towers. Structural systems that are arranged so that each member primarily resists axial forces only are usually known as trusses. Long slender members that are subjected to loading normal to their longitudinal axis must resist bending and shear forces and are called beams. A structural frame consists of members that must resist both: bending and axial forces. Finite element equations for analysis of trusses are obtained by a simple coordinate transformation of the two-node axial deformation element developed in Chapter 2. Elements for both plane trusses and three-dimensional space trusses are generated using this technique in the following two sections. The third section considers analysis of trusses subjected to initial strain due to temperature changes or fabrication errors. For modeling supports and various other constraints, it is often useful to introduce spring elements. The axial and torsional spring elements are presented in the fourth section. Beams are .governed by a fourth-order differential equation. Section 4.5 contains a detailed derivation of this equation, discussion of appropriate boundary conditions, and exact solutions for few simple cases. A two-node finite element for beam bending is developed in Section 4.6 using the Hermitian interpolation functions. For beams subjected to distributed loading, Section 4.7 describes a superposition procedure so that exact solutions for bending moment and shear force can be obtained using only one element per span. In Sections 4.8 and 4.9, the axial deformation and beam bending elements are combined together to develop elements suitable for analysis of general structural frames.
222
PLANE TRUSSES
4.1
PLANE TRUSSES
A truss is a structure in which members are arranged in such a way that they are subjected to axial loads only. The joints in trusses are considered pinned. Plane trusses, where all members are assumed to be in the x-y plane, are considered in this section. The general case of space trusses is considered in the following section. A plane truss element is an axial deformation element oriented arbitrarily in a twodimensional space. As.shown in Figure 4.1, in a local coordinate s that runs along its axis (0 :;; s :;; L), the element is exactly the same as the two-node axial deformation element developed in Chapter 2. Thus in terms of s the assumed displacement over the element is lI(S) = (
L- S
L
O:;;s:;;L
and the element equations in the local coordinate system are as follows:
Local degrees of freedom: Local applied nodal forces: where E = elastic modulus of the material, A = area of cross section of the element, L = length of the element, d l and d z are the displacements along the axis of the element, and PI and Pz are possible axial loads applied at the bar ends. It is assumed that there is no distributed axial load along the length of the element. Using these expressions equations for any truss element can be written. However, when there are several truss elements oriented arbitrarily, each element will have its own local axis and the direct assembly-procedure discussed in Chapter 1 will not work. To be able to Local element coordinates
/
Global coordinates dz
V2
Figure 4.1. Local and global coordinates for an axially loaded bar
223
224
TRUSSES, BEAMS,AND FRAMES
assemble element equations, we must refer all elements to one common reference coordinate system. Thus we define a global x-y coordinate system and locate all elements with respect to this system. The components of the axial displacements in the global coordinate system are the x displacements denoted by u and y displacements denoted by v. Each node thus has two degrees of freedom in the global coordinate system. The possible applied loads at the element ends are also decomposed into their x and y components. Thus in the global coordinates we have
Nodal degrees of freedom:
Applied nodal forces:
From Figure 4.1, it can easily be seen the vector d l is related to vectors Ll I
= d l cos a == dlls
VI
= d l cos(90 - a) == d l sin a == d.m,
Ll I
and VI as follows:
where Is is the cosine of the angle/between the element s axis and the global x axis and ms is the cosine of the angle between the element axis and the global y axis and are called the direction cosines. The angle is positive when measured from the positive x axis in the counterclockwise direction. Denoting the coordinates of the ends of the element by (xl' YI) and (x2' Y2)' we can determine the element length and the direction cosines as follows: I = cos a = x2 s
ms
= cos(90 -
a)
-
L
~I .
'
= sin a = Y2 L YI
Multiplying Ll I by Is and VI by ms and adding the two equations together give transformation from global to local degrees offreedom as follows:
The same relationship holds for the degrees of freedom at node 2. Thus the transformation between the global and the local degrees of freedom can be written as follows:
PLANE TRUSSES
From global to local:
From local to global:
Using this transformation, the element equations in the local coordinates can be related to the global coordinate system as follows: k.d,
=r
====> k.Td
= r,
Multiplying both sides by TT, we get
r,
Noting that TT is the transformation of applied loads from the local to the global coordinates, we have the following element equations in terms of global degrees of freedom and applied nodal loads in the global directions:
led = r where and
n
Carrying out matrix multiplication, we get the following element equations:
EA L
[
-1;
Is':ls P
Isms m;
-lm,
-Isms
-Isms -m;
z2s Isms
-Z;
-I,m,][ ] -m; Ll i
Isn~s m;
Vi
LIZ Vz
Ix
FI
= ;~ Zy
As discussed in Chapter 1, it is convenient to add the concentrated loads directly to the global equations at the start of the assembly. Thus in the element equations the right-handside load vector is taken as all zeros:
z2s EA L
Isms
-1;
Isms In;
-Isms -lsln s -m;
-1; -lslns
z2s Isms
z . -Ins -I,m,]["']
n
_ 0 LI?- -. 0 Vz 0
VI
Isins mZ s
These equations can be used to analyze any plane assemblage of bars in which individual elements can be assumed to resist axial loads. The assembly and solution procedures have
225
226
TRUSSES, BEAMS.AND FRAMES
been discussed in detail in Chapter 1. After computing nodal displacements for each element, the element solution is computed by first transforming the nodal displacements back to the local coordinates as follows:
Axial displacements:
The axial displacement at any point along the element can be computed as follows: u(s)= (
L- S
L
O::5S::5L
The axial strain is simply the first derivative of the axial displacement, giving constant strain over the element as follows:
The axial stress is a- = EE and axial force in the element is F = a-A. The sign convention used in these equations assumes that the tension is positive and the compression is negative.
Example 4.1 Six-Bar Truss Consider the simple six-bar pin-jointed structure shown in Figure 4.2. All members have the same cross-sectional area and are of the same material, E = 200 GPa and A = 0.001 m2 • The load P = 20 kN and acts at an angle a = 30°. The dimensions in meters are shown in the figure. Each, node in the model has two .displacement degrees of freedom and thus there are a total of ten degrees of freedom as shown in Figure 4.3. The displacements are identified by the letters u and v with a subscript indicating the corresponding node number. The calculations are similar to the truss examples presented in Chapter 1. Complete calculations for this example can be found on the book web site. The solution summary is as follows:
3 2.5
2 1.5 1
p
0.5
o
o
2
3
4
Figure 4.2. Six-bar truss
SPACE TRUSSES
8
6 5
7
10
9
"'--'1-------'........ 3
Figure 4.3. Six-bartrussnodaldegrees of freedom
+ MathematicalMATLAB Implementation 4.1 on the Book Web Site: Analysis of a plane truss 4.2
SPACE TRUSSES
By considering an axially loaded element to be arbitrarily oriented in a three-dimensional space, we can easily develop an element that can be used to find joint displacements and axial forces in any space pin-jointed framework. With the local s axis along the axis of the element, the local element equations are still the same as the two-node axial deformation element:
where E = elastic modulus of the material, A = area of cross section of the element, L = length of the element, d, and d z are the displacements along the axis of the element,
227
228
TRUSSES, BEAMS, AND FRAMES
Local element coordinates
Global coordinates
Figure 4.4. Local and global coordinates for an axially loaded bar in three dimensions
and PI and P2 are possible applied axial loads at the ends. In the global coordinates the nodal displacements and applied forces are as shown in Figure 4.4 and are as follows: III
VI
Nodal degrees of freedom:
d=
WI
Applied nodal forces:
r=
112
v2
w2 The transformation between the global to local degrees of freedom and vice versa is as follows:
where the direction cosines and the length can be computed from the nodal coordinates as follows: 1 = X2 -Xl. n = Z2 -Zl m = Y2 - Yl. S L S L' S L'
Using this transformation, the element equations in the local coordinates can be related to the global coordinate system as follows: kid,
=r, :::=? k,Td =r,
Multiplying both sides by TT, we get TTk/Td
= TT r/
SPACE TRUSSES
Noting that TTl'/ is the transformation of applied loads from the local to the global coordinates, we have the following element equations in terms of global degrees of freedom and applied nodal loads in the global directions:
kd
e
r
where and Carrying out matrix multiplication, we get the following element equations for a space truss element:
Assuming that the concentrated loads are added directly to the global equations at the start of the assembly, the element equations are
Ps EA L
mis nis,
-ps
msl s m; I1l sns
-mis
-mis -m;
-nis
-msns
nis msns
n;
-1;
-mis -nis -m/s -m; -msns -nis -msns -n;
-:nis -msns ·_n z s
ps
mis
ul VI WI
Uz Vz Wz
nis
mis
m;
»»,
nis
msns
12;
0 0 0 0 0 0
The assembly and solution procedures are standard and are discussed in detail in Chapter 1. After computing the nodal displacements, the element quantities can be computed in the usual manner. The axial displacement at any point along the element is computed by first transforming the global displacements into the local coordinates as follows: UI VI
Axial displacements:
o
~J :~
Vz
Wz
The axial displacement over the element is O::;;s::;;L
229
230
TRUSSES, BEAMS,AND FRAMES
The axial strain is simply the first derivative of the axial displacement, giving constant strain over the element as follows:
The axial stress is o: :::: EE and axial force in the element is F :::: erA. The sign convention used in these equations is that tension is positive and compression is negative.
Example 4.2 Three-Bar Truss Consider the simple three-bar pin-jointed structure shown in Figure 4.5. The cross-sectional area of elements 1 and 2 is 200 mm2 and that of element 3 is 600 mrrr'. All elements are made of the same material with E :::: 200 GPa. The applied load is P :::: 20 kN. The nodal coordinates are as follows: Node
x(m)
y(m)
z(m)
1 2 3 4
0.96 -1.44 0 0
1.92 1.44 0 0
0 0 0 2
Each node in the model has three displacement degrees of freedom, and thus there are a total of 12 degrees of freedom, as shown in Figure 4.5. The displacements are identified by the letters u, v, and w with a subscript indicating the corresponding node number. The calculation are similar to the plane truss examples presented in Chapter 1. Complete calculations for this example can be found on the web site. The solution summary is as follows: .
!
Nodal Solution
1 2 3 4
x Coordinate
y Coordinate
z Coordinate
Lt
v
w(mm)
960. -1440. 0 0
1920. 1440. 0 0
0 0 0 2000.
0 0 0 -0.178143
0 0 0 -2.46857
0 0 0 -0.367431
2. v2
2 U2
Figure 4.5. Three-bar space truss
. TEMPERATURE CHANGES AND INITIALSTRAINSIN TRUSSES
Element Solution
1 2 3
Stress (MPa)
Axial force (N)
101.873 66.0725 -38.5802
20374.6 13214.5 -23148.1
• MathematicalMATLAB Implementation 41.2 on the Book Web Site: Analysis of a space truss 4.3 TEMPERATURE CHANGES AND INITIAL STRAINS IN TRUSSES Stresses may be induced in indeterminate trusses due to temperature changes or because of forced fit during construction if a truss member is fabricated too short or too long. If an element of length L is fabricated too short by an amount M and is then stretched to make it fit during construction, the element is clearly subjected to the following initial strain:
M
EO
=L
Similarly, a temperature increase of IlT in a long slender bar causes a uniform expansion, in which case the element is subjected to the following initial axial strain:
=
EO
ost
where 0: is the coefficient of thermal expansion of the material. Thus in both situations an element is subjected to an initial strain before any extemalloads are applied. In order to account for these effects in trusses, we consider an axial deformation element with an initial strain EO' The axial strain obtained from differentiating the axial displacement u(x) is clearly the total strain. The stresses in the element will be caused by the difference in the total strains and the initial strains. That is, dU ) a:=E ( --EO x dx
The strain energy function for the element is now as follows:
u=~
J" EA(~; - r ~ .(2 EA(~;r EO
dx =
X2 dx-l.
EA ~;
2
EO
dx+
~ f,X EA(EO)2 dx
Following the derivation in Section 2.7 of Chapter 2, the assumed solution and its derivative are u(x)
= (N}
du(x) =(Nj dx
N2 )
C~):; NT d;
N~)(UI):;BTd; u2
N }
=
x-x'2.
L'
231
232
TRUSSES, BEAMS, AND FRAMES
The strain energy term can now be evaluated as follows:
or
where k is the usual stiffness matrix of an axial deformation element
..k =
l
X,
AE ( 1 AEBBT dx = -
L -1
XI
and Te is the equivalent nodal load vector due to initial strains
EO:
The last term in the strain energy involves the known initial strain and does not depend on the assumed solution. This term will drop out when writing the necessary conditions for the minimum of the potential energy. Thus the last term can be ignored, and therefore the presence of initial strains results in simply an equivalent noda1load vector Te• For analysis of plane and space trusses, this equivalentload vector is first transformed to the global coordinates using the appropriate tr~Jlsformation matrix:
For a plane truss,
For a space truss, -Is -Ins
T
=EAEO
-ns Is Ins
ns
SPRINGELEMENTS
l:~ L-
----'=_
o
16
(ft)
32
Figure 4.6. Plane truss
The equivalent nodal load vectors from all elements that are subjected to initial strain are assembled in the usual manner and global equations solved for nodal unknowns. The element stresses are obtained from the net strains:
.=E (dU dx - ) = E(B
o-x
EO
T
d-
EO)
=E (-d tL+ dz -
EO
)
Example 4.3 Plane Truss with Temperature Change Consider the five-bar pinjointed structure shown in Figure 4.6. All members have the same cross-sectional area A = in2 and are of the same material with E = 29,OOOksi and a = 6.5 x 10-61°P' The first element undergoes a temperature rise of lOO°F. The dimensions are shown in the figure. Complete calculations for this example can be found Onthe web site. The solution summary is as follows:
!
Nodal Solution 1 2 3 4
£I
v (in)
0 -0.0308148 0.0308148 0
0 -0.121333 -0.138667 0
Element Solution 1 2 3 4 5
Stress (ksi)
Axial force (kip)
-5.8179 -4.65432 -5.8179 -4.65432 3.49074
-2.90895 -2.32716 -2.90895 -2.32716 1.74537
4.4 SPRING ELEMENTS An axial deformation element can also be thought of as a liriear spring that is designed to resist axial tension or compression. The spring constant, usually denoted by k, is defined as the load that causes unit extension or compression in the spring. In the axial deformation element, if we assume Uz = 0 and £II = 1, then we have a unit extension of the bar, and
233
234
TRUSSES, BEAMS, AND FRAMES
Figure 4.7. An axial spring element
the finite element equations show that the axial load is PI = AE/L. Thus the term AE/L is equivalent to the spring constant k for a linear spring. Therefore, we can incorporate linear spring elements, such as the one shown in Figure 4.7, into a finite element model through the following element equations:
k( -11 For an arbitrarily oriented element these equations can be transformed to global coordinates using the same transformation matrices as those for the truss elements. Rotational or torsional spring elements can be defined in an analogous manner. Denoting the torsional spring constant by kT , the end rotations by 81 and 82 , and any applied twisting moments at the ends by My: ' My; , the equations for a torsional spring element are I 2 as follows: .,
Example 4.4 Find axial forces in the spring-bar assembly shown in shown in Figure 4.8. Use the following numerical values:
L = 30 in; F = 15,000 lb; Spring constant, k = 100,000 lb/in For the steel bar: E = 30 X 106 Ib/in"; A = 0.5 in2 For the aluminum bar: E = 107l1:J/in2 ; A = 1.2 in2 The assembly is modeled using one spring and two axial deformation elements as shown in Figure 4.9. There are three nodes, each with one displacement degree of freedom. Using pound-inches and following the usual steps, the solution is as follows: Specified nodal loads: Node
dof
Value
2
Lt:G
15000. Steel
, , - - - i l - - - -.... F
Aluminum L Figure 4.8. Spring-bar assembly
·1
SPRINGELEMENTS
Figure 4.9. Three-elementmodel for the spring-barassembly
Element 1:
100000 ( -100000
-100000)(U 100000
1)=(0) liZ 0
Element 2:
400000. -400000.) (liZ) ( -400000. 400000. u3
= (0)
500000. -500000.) (U z) ( -500000. 500000. u3
= (0)
0
Element 3:
-90000~·][::~1 = [15~001 900000. 0
100000 -100000 1. x 106 [ o -900000.
Global equations: -100000 Node Essential BC
~
dof
0
lI 3
Value
11
0
330 Global equations afterJIBC: 1. x 106 Uz
= 15000
Nodal solution: (uz ~ 0.015} The force in the spring is equal to the spring constant times its change in length: Force in spring
= k(uz -
u 1)
= 1500. lb (Tension)
The axial strain in the bars is Axial strain
= (u3 -
uz)IL
= -0.0005
The axial forces in each bar is equal to EAE, giving In the steel bar = -7500. Ib (Compression)
In the aluminum bar
= -6000. lb (Compression)
235
236
TRUSSES, BEAMS, AND FRAMES
4.5 TRANSVERSE DEFORMATION OF BEAMS A beam is a structural member that carries a load normal to its axis. The length of a beam is large as compared to its cross-sectional dimensions. In general, the cross section can be of any arbitrary shape and can vary along its length. However, the following discussion is limited to beams whose cross sections are symmetric with respect to the plane of bending. The x axis passes through the centroid of the cross section. Under these conditions, the beam is subjected only to bending and there are no axial or twisting forces.
4.5.1
Differential Eq~ation for Beam Bending
Consider a beam with a symmetric cross section that is subjected to load in the transverse direction, as shown in Figure 4.10. The moment of inertia is denoted by lex) and the mass . of the beam per unit length is denoted by m(x). This mass is due to self-weight and any additional weight distributed along the beam. The beam may be subjected to distributed load q(x, t) along its length. The governing differential equation can be written by considering the equilibrium of a differential element, as shown in Figure 4.11. The shear force and moment at x are denoted by V and M, respectively. Using the Taylor series expansion, these quantities at x + dx are V +(8V/8x)dx andM + (8M/8x)dx. The transverse displacement is denoted by vex, t). The acceleration is indicated by ii == 82v/8P. From Newton's second law of motion, a force, called the inertia force, is produced that is proportional to the mass of the element and acts in the direction opposite to the direction of motion. With the mass per unit length being m(x), the inertia force is mdxii. The only other force acting on the element is the applied transverse load q(x, t). Note that the sign convention adopted in drawing the free-body diagram in Figure 4.11 assumes that the internal moments causing compression on the top of the beam are positive. This means that along the left side of an element a clockwise moment is positive while on the right side a counterclockwise moment is positive. On the left side the shear force along the y axis is positive while on the right side the positive shear acts in the -y direction. As far as the external loading is concerned, an external applied moment in the counterclockwise direction is assumed positive while applied force is positive if it is acting along the +y axis. Considering summation of forces in the y direction, we have mdxii(x, t) == q(x, t)dx+ V - (V
+ ~~ dX) =* mii(x, t) +
~~
== q(x, t)
v
x
Figure 4.10. Beam bending
TRANSVERSE DEFORMATION OF BEAMS
qix.t)
U-LLJ \
r: aM
av
V+axdx
Figure 4.11. Forces acting on a differential beam element Summation of moments about the right-hand side gives
-q(x, t)dxdx/2-M - V dx +M +
aM dx = 0 -a x
Neglecting the higher order dY? term, we get
aM
v=ax In order to relate forces to deformations, a fundamental assumption in beam bending is that a plane section before bending remains plane and normal to the neutral axis after bending. This assumption implies that the axial deformation at a point y above the neutral axis is given by
dv u(x) = -y dx The axial strain
Ex
is
du
=-dx x
E
dZv =-y-z dx
For a linear elastic material the axial stress 0:.: and strain are related by Young's modulus E, giving
Thus the axial stress due to bending varies linearly over the cross section. The maximum compression is at the top and the maximum tension is at the bottom. Taking the moment of forces acting on a cross section, we get .
237
238
TRUSSES, BEAMS, AND FRAMES
L
where A is the area of the cross section and 1 = l dA is the moment of inertia or second moment of the area. Using this equation, we can relate moment and axial stress to get
~ = -EY(~~) =-~Y Differentiating the moment, we can relate the shear force to displacement as follows:
V= ax = !.-ax (E1 ax
a2v
aM
2
)
Substituting the derivative of V into the first equilibrium equation, the governing differential equation can be written as follows:
+ :~
mii(x, t)
(E1:~) =q(x, t)
This is a fourth-order partial differential equation. The equation involves a fourth-order derivative with respect to x and a second-order derivative with respect to t. We need four boundary conditions and two initial conditions for a proper solution. The beam boundary conditions are discussed in the following section. The initial conditions are the specified displacement and velocity along the beam at time t = O. V(x,O)
II
= Vo
where Vo and Vo are specified displacement and velocity values. For a static analysis situation, the inertia force is zero, and we have the following governing differential equation:
II
2 -2 (
a ax
II
'11
III
:II
a v)= E1ax 2
2
q(x)
'II
:11
/
4.5.2 Boundary Conditions for Beams The boundary conditions for beam bending involve specification of displacement v or any of its first three derivatives with respect to x. Some common beam boundary conditions are shown in Figure 4.12. Since the second derivative of displacement is related to the bending Simple Support
u=v=O
Roller Support
v=O
Fixed Support u=v e=o~
Internal Support
v=O
Internal Hinge
M=O
=~ Figure4.12. Typicalbeam boundary conditions
TRANSVERSE DEFORMATION OF BEAMS
moment and the third derivative to shear force, the boundary conditions at a point X o along a beam are as shown in the following table. Carefully note the convention used in showing these boundary conditions in the figures. Also note that the signs for internal moments and shears depend on whether we are considering a left side or a right sid~ of a section. Specified Quantity
Right Side of Section at X o
Left Side of Section at X o
Displacement
=v.tO
v(xo)
v(xo) =
o) - B BeX o) -= OV(X ox xO
) = ov(xo) - B Bexo-ox-xo
Rotation
V xO
=M.<0
Moment
_E/ 02V(xo)
= MxO
E/ 02v(XO)
Shear
E/ 03V(xo)
= F.<0
- E/ aT -
o.o?-
ox'
o.o?-
03V(Xo) -
F
xO
Boundary conditions at a free end: Bending moment:
a2 v
,
a~
v =~ ax
Shear force:
M=E/- =0'
[El2V]= a~
0
Boundary conditions at a fixed end: Deflection:
v
= 0;
Rotation:
B=
av =0 ax
Boundary conditions at a simple (pin) support: Deflection:
v = Q;
Bending moment:
Boundary conditions at a guided (sliding) support: Rotation:
av ax
B=·- =0'
'
v =~ ax
Shear force:
[Elax2V]= 2
0
More complicated boundary conditions are possible when beams are supported by other elastic members. Such supports are typically represented by extensional or rotational springs. The force in an extensional spring located at xo, with the spring constant k, is
This force must be balanced by the shear at the support. Using the free-body diagram shown in Figure 4.13 and the sign convention for shear at the right end of asegment, the appropriate boundary condition is Right-end extensional spring support:
~ [~l2V] _kv = 0
ax
a~
239
240
TRUSSES, BEAMS, AND FRAMES
-"1 Extensional spring support
Rotational spring support
O~~ i
Figure 4.13. Spring-supported boundary conditions
A rotational spring generates a moment equal to kB, and therefore the boundary condition at the right end of a beam supported by a rotational spring is . Right-end rotational spring support:
[Pv
EI aY} + kB = 0
From the free-body diagrams it is easy to see that, if the elastic supports are at the left end, the appropriate boundary conditions are as follows: Left-end extensional spring support:
II
IIII
Left-end rotational spring support:
I"'
,,'i: II'
4.5.3 Shear Stresses in Beams The basic beam theory does not take into account the deformations due to shear forces. Thus the shear stress does not enter into the governing differential equation. However, knowing the shear forces, we can compute the corresponding shear stresses using the following equation from the mechanics of deformable bodies: VQ
T=-
It
where V is the shear force, I is the moment of inertia, and Q and t are related to the point in the cross section where the shear stress in being computed, t being the width of the cross section at this point and Q the moment of the area above this point about the neutral axis. See Example 4.5 for an illustration. .
4.5.4
Potential Energy for Beam Bending
For a linear elastic beam the axial strain and stress due to bending are
TRANSVERSE DEFORMATION OF BEAMS
Thus the strain energy for a beam element with end coordinates at X o and Xl can be written a& follows: '
u= ~
1
CTxExdV =
~ L:'1EY(~~)Y(~~)dAdX = ~ L<~ El(~:~r dx
The potential of the applied loads is equal to the negative of the work done by the external applied forces:
w = t" qvdx+ Lpiv(x) Jxo where v(x) is the transverse displacement at the location of the concentrated applied load Pi' Thus the potential energy for beam bending can be written as follows:
. II = U - W
2 = -1 LX, E1 (d V)2dx- LX, qvdx- Lpiv(Xi) 2
-?
xo
dx'
X
o
Using calculus of variations, it is possible to show that a function v(x) that minimizes the potential energy is a solution of the differential equation governing the bending of beams.
4.5.5 Transverse Deformation of a Uniform Beam Consider a uniform beam simply supported at the left end and spring supported at the right end, as shown in Figure 4.14. The beam is subjected to a linearly increasing load q(x) = ax!L, where a is a given constant. The problem is described in terms of the following boundary value problem: E1
d4v _ Cl.,,,(. dx4 - L.'
v(O) = 0; .
E1
0
V El2d~ (0) = 0
d 3v (L ) _ kv(L) -dx 3
=O' .,
E1
d2v
(L )
d~
=0
y q(x) = a x(L
k
EI
L
-I
Figure4.14. Uniform beam subjected to a linearly increasing load
241
242
TRUSSES, BEAMS,AND FRAMES
An exact solution of the problem can be obtained by integrating the differential equation four times and then using the boundary conditions to evaluate the resulting integration constants. Integrating both sides of the differential equation, we get d3
v ax
2
El dX3
-
d 2v El dx 2
-
2L
ax3
dv
= C1
6L
= cjx+ C2
ax 4
x2
El dx - 24L = c1 "2 + C2X + C3
X3
axS Elv-
where
C1>""
C4
.
are integration constants. Using the boundary conditions, v(O)
Thus the exact solution of the problem is _ ax(120EIL + k(7L 4 - 10L2x2 + 3x 4)) ( 360ElkL v x) -
As k
-7 00,
veX)
the solution approaches that of a simply supported beam:
= li
k->~
(ax 0 20E IL + k(7L4 - lOL2x2 + 3X4))) = ax(7L4 .
360ElkL
2L2 lOx + 30) 360EIL
Assuming El = 1, L = 1, and a = 1, the solution for various k values is as shown in Figure 4.15. With increasing k the spring-supported end displacement is getting small, as expected.
4.5.6
Transverse Deformation of a Tapered Beam Fixed at Both Ends
Consider a nonuniform beam in which the moment of inertia varies linearly along the span, lex) = 100 + b:x/L), where 10 is a reference moment of inertia and b is a constant. The beam
TRANSVERSE DEFORMATION OF BEAMS
vex) 0.008 0.006 0.004 0.002 k=500
+ __~ _ ~__~__~_~k=lOOO 0.2
0.6
0.4
I
0.8
Figure 4.15. Exact solutions for spring-supported beam with various spring constants
is fixed at both ends, as shown in Figure 4.16. The beam is subjected to a linearly increasing load q(x) = ax/L, where a is a given constant. The problem is described in terms of the following boundary value problem:
2
~ (EI(X)d V) = 2 d~
dx
EI(x) = Elo v(O) = 0;
ax.
L'
0
(1 + ~) dv(O) = 0
dx v(L)
= 0;
dv(L)
dx
=0
The actual derivation of the following solution is quite tedious. However, by direct substitution, it can be verified that the solution satisfies the governing differential equation:
y
L
-j
Figure 4.16. Nonuniform beam subjected to a linearly increasing load
where C1"'" C4 are integration constants. Using the boundary conditions, the constants can be evaluated and the solution expressed as follows: IIII III "ll
vex)
= (a((L -
x)2(2( -1
+ b)L 2 + 2(-2 + b + bZ)Lx + b(2 + b)Y}) log[L]
+xz(-6Lz + 3b2Lz + 4Lx+ 2bLx- 2bx 2 - bZY})log[(1
1111
IIll!I'"I:
-
Ill!
1::1
I",
2(bL3x
- bZL3x - 3bL zy}
+ bL4log[L + bx] -
Il!'
Ih.
(72b z(2b
+ 2b zL2x2 + 2bLx 3
-
+ b)L] L41og[L + bx]
bL 3xlog[L + bx] + bZL3xlog[L + bx])))/
+ (2 + b) log[.L] - (2 + b) log[(1 + b)LDE1 0 )
Figure 4.17 shows a plot of this solution with (b = 0.1, a
4.6
-
bZx4
= 1, L = 10, E10 =
I}.
TWO-NODE BEAM ELEMENT
Recall from the discussion in Chapter 2 that for a fourth-order problem the essential boundary conditions are the solution and its first derivative. For beam bending, therefore, both the transverse displacement v and rotation == dv/dx are essential boundary conditions. In order to be able to satisfy these boundary conditions during the assembly and solution phase, we must choose both v and as degrees of freedom for each node. Thus the simplest two-node beam element has four degrees of freedom, as shown in Figure 4.18. For simplicity in integrations the moment of inertia and the distributed load are assumed constant over each element. Furthermore, concentrated loads FI , Fz and moments M I , Mz are allowed only at the ends of an element. As demonstrated through examples later in the section, even this simple element is enough to model most practical beam problems. For uniform beams only one element per
e
e
TWO-NODE BEAM ELEMENT
V2
q
x Xl
S=o
s=L L Figure 4.18. A two-node element for beam bending
span is needed to get exact solutions. For concentrated loads along the span or changes in section dimensions, one simply needs to start a new element at the locations where such situations occur. Even beams made up of different materials can be modeled, simply by starting and ending elements at the material interfaces. Only for variable-cross-section beams is it necessary to use a large number of constant moment-of-inertia elements to get a good solution.
4.6.1
Cubic Assumed Solution
Since both displacement and rotation are used as degrees of freedom for each node, the appropriate assumed solution is written using the Hermite interpolation. With the general formula given in Chapter 2 it is possible to write shape functions for an element with arbitrary end coordinates xI and x2' However, the resulting expressions are very messy. More convenient expressions are obtained by introducing the following change in coordinates: O::;,s::;,L
ds
= dx
dv ==} -
dx
=-dv ds
The four interpolation functions in terms of s can easily be written as follows: Data point locations: {O, L} Lagrange interpolation functions for these points:
Derivatives of Lagrange interpolation functions:
245
246
TRUSSES, BEAMS, AND FRAMES
Derivatives at the given points:
Using these, the Hermite interpolations for data values are
and those for derivative values are
Vector of Hermite interpolation functions:
NT
= {2S 3 _ L3
2
3s + 1 L2
'L2
L
3
2
2
.::=. _ 2s + S 3s
'L2
_
2s s3 _:::} L 3 'L2 L
Thus the assumed solution is
III 111I
1111
v(s)
= (1$' L'
_
2
3s2 L
+1
III! m:
Ih' Ill'
::11 ,I
4.6.2
Element Equations Using Rayleigh-Ritz Method
The potential energy for the beam element is as follows:
or
Differentiating the assumed solution twice with respect to s, we have
TWO-NODE BEAM ELEMENT
Squaring this (making sure that vectors d are on the outside),
where
BB
T
=
36(L-2s)2 -L-6-
12(2L-3s)(L-2s) L'
12(2L-3s)(L-2s) L'
4(1£-3s)2 L4
36(L-2s)2 12(2L-3s)(L-2s) L'
12(2L-3s)(L-2s) L'
_36(L-2s)2 L6
4(L-3s)(2L-3s) L' 12(L-3s)(L-2s)
36(L-2s1'
~
IJ
12(L-3s)(L-2s) L'
4(L-3s)(2L-3s) L4
12(L-3s)(L-2s) L'
12(L-3s)(L-2s) L'
-~
4(L-3d -L-'-
Using this, the strain energy term can be written as follows:
where
Ia
L
(
12(2L-~~)(L-2S») ds
rL (4(2L-3S)2) d L4 S
Jo
".
"'J
Carrying out integrations, we get
12 k _..
= El
6L L 3 [ -12 6L
6L 4L2 -6L 2L2
-12 -6L
12 -6L
The work done by the distributed load can be evaluated as follows:
L
W
q
= Jor
qv ds
r
= Jo
qNTd ds
= rTq ddTrq
where
t(l- ~+ ZS)ds
r
L 2s' + 'jJ s3 ) ds Jo ( s - T
rL (3s' Jo L2 L
r Jo
(
-:-
s' - L
2s3) L3 ds s3 ) ds + 'jJ
=
q;[jJ
247
248
TRUSSES, BEAMS, AND FRAMES
Equivalent nodal loads
Distributed load
qL/2
q
LLillLJ
qL2/12
.---k po L
f-'-L-...j
qL/2
'::~:~::
.---k. qL
2/12
.:".: : :d "
f-'-L-...j
Figure 4.19. Equivalent nodal loads due to a uniformly distributed load on an element
Interpreting tenus in the vector rq , we see that a uniformly distributed load on an element is actually applied as' nodal loads qV2 and moments qL2/l2. With the sign convention being used, the equivalent nodal values are as shown in Figure 4.19. The work done by the concentrated loads is
The potential energy can now be written as follows:
Ilill 11111 111111
III!I 1"
The necessary conditions for the minimum of the potential energy give
1,.
III::: hUH ~l!!t
::111
J,,,..
ijljll'
Willi
Thus the beam element equations are as follows:
-12 -6L
12 -6L
Assuming that the concentrated loads are added directly to the global equations at the start of the assembly, the element equations are .
12 EI 6L L 3 [ -12 6L
6L 4L 2 -6L 2L2
-12
-qL 12 -6L
For a given beam problem, the element equations can be assembled and nodal displacements can be obtained in the usual manner. Once the nodal displacements are known, the
TWO·NODE BEAM ELEMENT
complete finite element solution can be obtained from the interpolation functions:
The bending moments and shear forces can be obtained by differentiating the displacement and multiplying by E1: d2v di'
3v
M(s) =E1-'
Yes)
d =E13 ds
If desired, the axial' and shear stresses can be computed from the bending moment and shear forces by using the equations derived earlier. For a symmetric section with height h the maximum stresses due to a given moment and shear are as follows: Maximum bending stress: Maximum shear stress:
M(h/2)
O:"max T max
= --1VQmax = -1-t-
where Q max is the moment of half of the area about the neutral axis and t is the width of the section at the neutral axis. Example 4.5 Find the deflection, bending moment, and shear force distribution in the three-span continuous beam shown in Figure 4.20. The point load is acting at the center of the middle span. Use the following numerical data: L = 20ft;
F = 20,000 lb;
Section: W18 x 40
The W18 x 40 section is a ~t.qndard steel I-shape section with the moment of inertia 1 612 in" and the dimensions as shown in the figure.
---<>j<>---
h = 17.9 br = 6.015
L
-I-
L
--l
r
w O>315J . . [t.·. tr = 0.525 1"
t. =
tr
.....br<
:
Figure 4.20. Three-span beam with I-shape cross section
=
249
250
TRUSSES, BEAMS, AND FRAMES
FJ2
~::c,,",:,c==.;;;-~ I-
L
~/2-----<>j
Figure 4.21. Two-element model using symmetry
For maximum shear stress, Q is the moment of the area of half of the l-section about the center. Using the given dimensions, we get Q
= ~ (~- tf)tlV(~ -tf) + (~- i)blJ = 38.6135 in3
Taking advantage of symmetry, we need to model only half of the beam as shown in Figure 4.21. Along the line of symmetry there is unknown vertical displacement but the rotation must be zero. Thus this boundary condition can be represented as a guided support. Also, only half of the load is carried by the symmetric half. The simplest model is a two-beam element model as shown in the figure, Use kips (1 kip = 1000 lb) and inthes. The displacements will be in inches, moments in kip = inches, and stresses in ksi. Specified nodal loads: Node
Adding element equations into appropriate locations, we have 1848.75 15.4063 295800. 1848.75 -15.4063 -184~.75 147900. 1848.75
o o
o
o
-15.40~3
-1848.75 138.656 5546.25 -123.25 7395.
o o 1848.75 o o 147900. 7395. 5546.25 -123.25 295800. -7395. 887400. 123.25 -7395. -7395. 591600. -7395. 295800.
Essential boundary conditions: Node dof Value 1 2
v2
0 0
3
83
0
VI
Remove (1, 3, 6) rows and columns: 295800. 147900. 0 J[8 IJ 147900. 887400. -7395. 82 [ o -7395. 123.25 v3
=[
0 J 0 -10.
Solving the final system of global equations, we get (81
= 0.000811359, 82 = -0.00162272, "s = -0.178499)
Solution for element 1: Left node: XI = 0; Right node: x 2 = 240. Local coordinate: s = X -xI = x; Length, L = 240. Using these sand L values, the interpolation functions in terms of x are NT :::;; .(1.44676 X 10- 7~ - 0.0000520833;J
0.0000173611~ - 0.00833333~ +x, 0.0000520833~ - 1.44676 x 1O-7~,
0.0000173611~ - 0.00416667~)
Nodal values: aT vex) = NT
= (0,0.000811359,0, -0.00162272)
a = 0.000811359x -
1.40861 x 1O-8x3 E = 29000; 1= 612; h = 17.9; Q = 38.6135; t = 0.315
= EI d 2v/dx 2 = -1.5x Vex) = dM/dx = -1.5 M(x)
=M(hl2)11 = -0.0219363x T = VQ/(It) = -0.300447
tr
+ 1,
o
o o o
-10
o
252
TRUSSES, BEAMS, AND FRAMES
V (kip) 10
v (in) Displacement 0.05 120
-0.05 -0.1 -0.15
M (k-in) 800 600 400 200
Shear force
8 6
24
4 2
£T (ksi) Bending stress 12.5 10 7.5 5 2.5
Moment
-2.5
-200
~~~
-5
Figure 4.22. Solution of three-span beam
Solution for element 2: Left node: xl = 240.; Right node: x2 = 360. Local coordinate: s =X - XI = X - 240.; Length, L = 120. Using these sand L values, the interpolation functions in terms of x are NT
0.0000694444(x - 240.)3 - 0.00833333(x - 240.)2j Nodal values: d T = (O, -0.00162272, -0.178499, OJ vex) = NT d = 9.39073 X 1O-8x3 - 0.0000777552x2 + 0.0194726x - 1.4929 E = 29000; 1 = 612; h = 17.9; Q = 38.6135; t M(x) = E1 d 2v/dx2 = 10.x - 2760.
= 0.315
Vex) = dM/dx = 10. o: = M(hl2)/1 = 0.146242x - 40.3627 T = VQ/(1t) = 2.00298 The deflection, shear force, bending moment, and bending stress diagrams are shown in Figure 4.22. From direct integration of the governing differential equation, it can easily be verified that this is the exact solution. Thus there is no need to use a refined model. Example 4.6 An overhanging beam is connected to a cantilever beam through a simple pin connection as shown in Figure 4.23. Analyze the beam if the right support settles downward by 10 mm. Use L = 3 m and E1 = 180 MN . m2.
TWO-NODEBEAM ELEMENT
l
EI . / :::.::,.. :::: :--.-.. _..;;Z;;;;.
-I-
,I·
L
L---l
Figure 4.23. Beams connected through a pin connection
Constraint : v 3 =v 4 vI
8
1
V3V4
~~---=--21( .
83
V5
3 84
5
8
~
Figure 4.24. Finite element model of beams connected through a pin connection
A three-element model is adequate for this beam. Modeling the internal pin connection requires special consideration because the rotations at either side of the pin are independent. Placing only a single node at this location and using the usual assembly process will imply continuity of rotations across the pin, which obviously is not the behavior of a pin connection. A way to model the pin is to place two separate nodes at the pin location and use a multipoint constraint to link the displacements at these two nodes while keeping the rotations independent. Thus the finite element model consists of five nodes and three elements as shown in Figure 4.24. Nodes 2 and 3 are physically at the same location. Element 2 is connected to nodes 2 and 3 while element 3 is connected to nodes 4 and 5. We use meganewton-millimeters units for numerical calculations. Substituting given numerical values into the beam element equations, we have the following element equations:
The known boundary conditions are VI == ()I .== v2 == 0, "s == -10 rom, and ()s == O. The zero boundary conditions are incorporated by simply removing corresponding rows and columns to get
Setting "s == -10 means we remove the last row and move -10 times the last column to the right-hand side. Thus the final system of equations is as follows:
The solution for nodal unknowns can now be obtained by using the Lagrange multiplier method to impose the multipoint constraint as follows:
TWO-NODE BEAM ELEMENT
V(X), mm
+-==-"'""--...:.--2L
L
-2
M(x),MN'mm X
3L
200 100,
-4 -6 -8 -10
;f~"
-100 -200 Figure 4.25. Displacement and bendingmomentplots
Augmented system of equations: 480000
-120
120000
0
0
0
-120
2
-120
0
0
1
240000
0
0
0
4
240
-1
480000 0
0 0
120000
25 -120
0
0
0
0 0
0 1
0 0
25 240 -1
82 v3 83 v4 84 it
0 0 0
=
8
-5
-2400 0
Solution:
The following complete solution over elements can be computed in the usual way. Note that the magnitude of the Lagrange multiplier is equal to the shear force at the pin. This is no coincidence. In structural problems a Lagrange multiplier is interpreted as the force necessary to impose the constraint. Here the constraint is on the transverse displacement across the pin. Thus the Lagrange multiplier will have an interpretation of force to maintain the transverse displacement continuity, This force is exactly the shear force at this location, and hence the value of the Lagrange multiplier is equal to the shear force. If a constraint involved rotations, the corresponding Lagrange multiplier will be equal to the moment at that location. Plots of displacement and bending moment are shown in Figure 4.25. The displacement plot clearly shows that the slope is not continuous across the pin and hence the internal hinge is modeled correctly: Displacement, vex)
Moment, M(x)
1
x2 2700000 -
400
2<
2
4x
12150000000 -
45 -
1600 -3-
45
3
4x
24300000000 -
1600 . -3-
45
x3
3 - IS
8100000000
x3 x3
x3 675000
i'
+
1350000
x 180 x
+ 300
50
-
"9 10
45 -
Shear, V 2
-IS 4
4
Example 4.7 Find the deflection, bending moment, and shear force distribution in the nonuniform beam shown in Figure 4.26. Use the following numerical data: L=2m;
F
= 18kN;
q = lOkN/m;
E = 210 GPa;
255
256
TRUSSES, BEAMS, AND FRAMES
F
q
1::;;;:;~:;;";'';;:;=;~1~~:~:~,:,Ll.1··jEJ-l·bL r-- L
-I-
L
-I-
VI
V2
V3
el~
e2~
2 e3~
L----J V4
3
e4~
1
3
5
7
2t
4~
6~
8~
Figure4.26. Nonuniform beam and three-element model
Placing a node at the concentrated load location and at the location of change in cross section, we have the three-element model shown in the figure. Using kilonewtons and meters, the solution is as follows: Specified nodal loads: Node
Solution for element 1: Left node: Xl = 0; Rig~~ node: x 2 = 2. Local coordinate: s = X - Xl = X; Length, L = 2. Using these sand L values, the interpolation functions in terms of X are NT
-5 Figure 4.27. Three-element solution of nonuniform beam
Solution for element 3: vex) = NTd = -0.0000146684X3 + 0.000283872x2 - 0.00155329x + 0.00226871 M(x) = EI d 2v/dx2 =47.6905 ~ 7.39286x Vex) = dM/dx = -7.39286 The deflection, bending moment, and shear force diagrams are shown in Figure 4.27. The exact solution in the distributed load segment should show a linear shear diagram and a quadratic bending moment diagram. However, since the element is based on a cubic interpolation, it gives a constant shear and a linear moment diagram, and therefore the solution for moments and shears is not very good. To get accurate results, we could use more elements in the segment with the distributed load. Using five elements in the distributed load segment, we get the moment and shear force diagrams shown in Figure 4.28 that are clearly a lot closer to the expected solution. However, for uniform beams it is not necessary to use more than one element per span. A simple superposition procedure is presented in the following section that gives exact solution without having to use many elements in the loaded span. That procedure obviously is preferred for practical applications. M (kN -m~
V (kN) 20 15
Moment
10
x -10 -20 -30
Shear force
10 5
-5
-10 -15
Figure 4.28. Solution with five elements in the distributed load segment
x
UNIFORM BEAMS SUBJECTED TO DISTRIBUTED LOADS
4.i'
UNIFORM BEAMS SUBJECTED TO DISTRIBUTED LOADS
As seen in Example 4.7, the simple two-node beam element does not give correct moments and shears for elements with distributed loads. The standard finite element approach for improving a solution is to use more elements. As demonstrated in the previous example, the approach works, and as. we increase the number of elements in the loaded span, the results do get closer to the exact solution. However, for uniform beams there is a much simpler alternative based on the principle of superposition. The idea exploits the fact that the two-node beam element, based on the cubic interpolation functions, gives exact solution for uniform beams subjected to concentrated loads. This was demonstrated in Example 4.5. The finite element formulation converts the distributed load to equivalent loads and moments at the nodes. The finite element solution is therefore exact for these equivalent concentrated loads and moments. The approximations are thus limited to the span that is loaded with the distributed load. This observation suggests a two-step solution procedure. In the first step we follow the usual finite element procedure, with only one element even for spans with distributed loads. In the second step the span with the distributed load is isolated and analyzed separately.' Both ends of this isolated span are assigned fixed-end boundary conditions so that the solution will not affect the displacements outside of this span. This is known as the fixed-end beam solution. The final solution is the superposition of the finite element solution and the fixed-end beam solutions for spans with distributed loads. In actual calculations, we do not even have to think of the process as a two-step process. We simply create a finite element model based on the usual considerations of supports, changes in the cross section, and concentrated load locations. After solving for the nodal displacements in the usual way, we compute the displacement solution over an element from the following equation:
v(s) = (
z:; - ~ + 1
where vf is a fixed-end beam solution due to a distributed load on an element. In the following sections we derive fixed-end beam solutions for uniform and trapezoidal loading and illustrate the process of computing element solutions, Fixed-End Beam Solution with Uniform Loading For a fixed-end beam with uniformly distributed load q, shown in Figure 4.29, the exact solution can easily be obtained by direct integration of the governing differential equation as follows:
Governing differential equation:
d4v EI-4 -q=O; ds
Boundary conditions:
v(O)
=0;
dv(O)
dx
O
= 0;
veL)
= 0;
dv(L)
dx
=0
259
260
TRUSSES, BEAMS,AND FRAMES
q
s=O
s=L
Figure 4.29. A fixed-end beam subjected to uniformly distributed load
Integrating four times and introducing four integration constants c 1, ••• , c4 ' we have
Using the four boundary conditions, we have
Solving these equations, the integration constants are
Thus the exact solution for a fixed-end beam with uniformly distributed loading is as follows:
~tCs) =
q(L - s)2s2
24EI
The subscript f is used to indicate that this is the fixed-end beam solution.
Example4.8 For the beam in Example 4.7 use superposition with only three elements to find the exact solution for the deflection, moment, and shear in the beam. The first step is to perform a standard finite element analysis. The equivalent nodal loads on the finite element model are shown in Figure 4.30. As shown through detailed calculations in Example 4.7, the nodal solution is as follows:
CompleteTableof Nodal Values 1 2 3 4
v
e
0 -0.000213152 -0.00034127 0
o -0.000131378 0.0000136054 0.000268991
Using the two-node beam interpolation functions, the following displacements were also computed in Example 4.7 for the three elements:
UNIFORM BEAMSSUBJECTED TO DISTRIBUTED LOADS
.•.
~I
L-<>J
Figure 4.30. Equivalent concentrated loading on the finite element model
Solution for element 1: VeX)
=NTd = 0.0000204436.:2 -
0.0000941752x2
Solution for element 2: VeX)
= NTd = 2.58645 X 10-6.:2 + 0.0000129677~ -
0.000214286x + 0.000142857
Solution for element 3: VeX)
=NTd = -0.0000146684.:2 + 0.000283872x2 -
0.00155329x + 0.00226871
For the first two elements these are the exact solutions. For the third element, in order to obtain the exact solution, we must add the fixed-end beam solution. Substituting the numerical data, the fixed-end solution for this span is obtained as follows:
For the third element: Left node: xl = 4.; Right node: :~2 = 6. Local coordinate: s = x - xl = X - 4.; Length, L = 2. Thus
Therefore the exact solution for this element is VeX) = NTd + Vj(x) = -0.0000146684.:2 + 0.00028387~ - 0.00155329x
+ 0.00226871 + ( -
(6-X)2(X-4)2) 201600
giving VeX)
= -4.96032 X 1O-6x4 + 0.000084538x3 + 0.000827664x - 0.000588435
0.000450255~
261
262
TRUSSES, BEAMS, AND FRAMES
M (kN -m)
Moment
V (kN)
Shear force
20
10
15 10 5
-10
x
-5 -10 -15
-20 -30
Figure 4.31. Exact three-element solution for beam with distributed load
This is a fourth-order solution and hence will result in a quadratic bending moment diagram and linear shear force: . M(x) :::El d 2v/dX2 ::: -5.x2 + 42.607lx -75.6429
Vex) ::: El d3v/dX: ::: 42.6071 - 1O.x The bending moment and shear force diagrams are shown in Figure 4.31. They clearly show that the solution is exact.
• MathematicafMATLA'B Implementation 4.3 on the Book Web Site: Analysis of beams Fixed-End Beam Solution with Trapezoidal Loading In examples so far only the uniformly distributed beam loading has been considered. However, any other loading can easily be handled in exactly the same manner. We must derive an equivalent load vector to use in the finite element analysis. Also we must obtain an exact fixed-end beam solution for use in the superposition to get exact solutions for loaded spans. Consider a beam element with fuad varying linearly from a value qj at the first node of the element to a value q2 at the second node, as shown in Figure 4.32. Using linear Lagrange interpolation functions, this loading can be expressed as follows: q(s) ::: sq2 _ (s - L)ql L L
Using the beam interpolation functions and the above q(s), the equivalent load vector can be obtained as follows: L Jor q(s)( 1 - 3s' L'
q
Jro Nq ds
;}oL(7qj + 3q3)
r L q(s)(s - T 2s'- + "iJ i3) ds Jo
L
r :::
) + 2$3 L3 ds
>
fo L2(3ql + 2q2) :::
L
r q(s)(3s'Jo
v - u2$3) ds IoL q(s)( - f: + f,:) ds
;}oL(3ql + 7q2) -foL2(2ql + 3q2)
Interpreting terms in the vector rq , we see that a trapezoidal loading on an element is actually applied as nodal loads and moments, as shown in Figure 4.32.
UNIFORM BEAMS SUBJECTED TO DISTRIBUTED LOADS
Distributed load
Equivalent nodal loads
Figure 4.32. Equivalent nodal loads due to a trapezoidal distributed load on an element
For use in superposition, the fixed-end solution with trapezoidal distributed load q(s) can be obtained by solving the governing differential equation as follows: Governing differential equation:
d 4v EI-4
Boundary conditions:
v(O)
- q(s)
ds
= 0;
= 0;
O
dv(O) =
dx
O. '
v(L)
= 0;
dv(L)
=0
dx
Integrating four times and introducing four integration constants c 1, ... , c4 ' we have
Using the four boundary conditions, we have
Solving these equations, the integration constants are C2
= 0;
C - _ 3 -
-3q I L2 - 2q2 L2. l20EI
' .
Thus the fixed-end beam solution for trapezoidal distributed loading is as follows:
The subscript f is used to indicate that this is the fixed-end beam solution. Example 4.9 Consider a uniform beam subjected to triangular loading. The beam is simply supported at left end and is spring supported at the right end, as shown in Figure 4.33.
263
264
TRUSSES, BEAMS,AND FRAMES
y
q(x)
=a xfL x k
,.
L
·1 Vz
VI
61C
6z
cJ
Figure 4.33. Uniform beam subjected to a linearly increasing load
Use superposition to get the exact solution using only one element. Assume the following numerical data: a = 5kN/m;
L=5m;
E
=200 GPa;
k= 275N/mm
Use kilonewton-meters: k = 275 leNim For the given triangular load ql as follows:
= 0 and qz = 5. Thus the equivalent nodal load vector is /
(7ql + 3qz)L 20 (3q I
= (7 x 0 + 3 x 5)5 = 3 75 kN. 20
(3ql + 7qz)L = 8.75 leN 20
.,
:~qz)Lz =4.16667 kN . m
(2ql + 3qz)Lz = -6 ?5 kN . .m 60
The equations for element 1 are as follows: Element ends
The specified nodal spring stiffnesses are as follows: Node
2
dof
Value 275
=5.
=[3.75 4.16667 ] 8.75 -6.25
UNIFORM BEAMS SUBJECTED TO DISTRIBUTED LOADS
The spring contributes to the v2 degree of freedom. Adding the spring constant to the diagonal term of the third row, we have the global equations as follows:
Solving the final system of global equations, we get
{8 1 = 0.0668245, v2
= 0.030303,82 = -0.0633838}
The solution for element 1 is as follows: Left node: XI = 0; Right node: x2 = 5. Local coordinate: s = X - XI =X; Length, L = 5. Using these sand L values, the interpolation functions in terms of X are NT = {O.016x 3
-
0.1~
+ 1, 0.04~ -
0.4~
+ x, 0.1~ -
0.016~, 0.04x 3
-
O.~}
Nodal values: aT = {O, 0.0668245, 0.030303, -0.0633838} vex) = NT a = -0.00034722Y e- 0.0104167x2 + 0.0668245x Fixed-end beam displacement, vj(x) = 0.0000416667(5. - xfx2(x + 10.) Adding this value, the complete element solution is Vex) = 0.QQ00416667~ - 0.00347222~ + 0.0668245x M(x)
=EI d2V/d~ = 0.166667~ -
4. 16667x
Vex) = dM/dx = 0.5~ - 4.16667
The bending moment and shear force diagrams are shown in Figure 4.34. An exact analytical solution for the problem was presented in an earlier example. Substituting the given numerical data into the analytical solution, we get Vex)
= ax(l20EIL + k(7L 4 -
lOL 360ElkL
2x2
+ 3x4 )
.
= 0.0000416667~ - 0.0034722~ + 0.0668245x
which is exactly the same as that obtained using the finite element analysis procedure with superposition.
265
266
TRUSSES, BEAMS,AND FRAMES
M (leN-m) Moment
V (leN) 8 6 4 2
x
-2
-4 -6
Shear force
x
-2
-8
-4 Figure 4.34. One-elementsolution
4.8
PLANE FRAMES.
Members in a plane frame are designed to resist axial and bending deformations. The twonode beam element and the axial deformation element are combined together to form an element that can be used to analyze rigid-jointed planar frameworks. It is assumed that the axial and bending effects are uncoupled from each other, which is a reasonable assumption within the framework of small-deformation theory. As shown in Figure 4.35, a local coordinatesystem is established for each element. In this local coordinate system the s axis is along the element axis, with positive direction being from node I to node 2. The local t axis is 90° counterclockwise from the saxis. Each node has three degrees of freedom-two translations and one rotation. The moment of inertia and the distributed transverse load are assumed constant over each element. Furthermore, concentrated loads are allowed only at the ends of an element. The distributed load qs is in the axial direction and qt is in the transverse direction. These applied loads on an element are positive if they act in the positive directions of the local axes. In the local coordinate system, the element equations are simply a combination of the axial deformation element and the beam element. With the order of degrees of freedom as shown in Figure 4.35 the local element equations are therefore as follows: T
EA
0
0
IF
0 -T
IF 0
0
-IF
0
6E/
EA
12E/ 6E/
12E/
L2
EA
0
-T
0
6E/
0
-IF
4E/
0
-IF
0
EA
T
0
0
-IF
0
IF
12E/
-IF
2E/
0
-IF
IF L
6E/
L
0
12E/
6E/
IF
6E/
2E/
L
6E/
6E/
4E/
dt d2 d3 d4 d5 d6
~qsL ~qtL I L2 Uqt
~qsL ~qtL
::=}
k, dT == TT
I L2 -uqt
L
Assuming a common global x-y reference coordinate system for all elements in the frame, the degrees of freedom in the global directions are denoted by u I , vI' 81, £1 2 , V2' and 82 , To develop the transformation between the global and the local degrees of freedom, we can see from Figure 4.35 that the vector d t is related to vectors u l and VI as follows: UI
== d l cos e == dlls
VI
== d l cos(90 - a) == d, sin « == dims
PLANE FRAMES
Global coordinates
Local element coordinates
d~~)/d4
V2
2
./,~'-+.
y
Figure 4.35. Plane frameelement
where Is is the cosine of the angle between the element s axis and the global x axis and Ins is the cosine of the angle between the element axis and the global y axis and are called the direction cosines. The angle is positive when measured from the positive x axis in the counterclockwise direction. Denoting the coordinates of the ends ofthe element by (Xl' YI) and (xz' yz), we can determine the element length and the direction cosines as follows: I = cos c =
X -x _Z_ _ l .
s
ms = cos(90 - a) = sin e =
L
'
"t" y -y
Multiplying £l l by Is and vI by ms and adding the two equations together give transformation from global to local degrees offreedom as follows:
Similarly, £l l
= :':'dz cos(90 -
a) == -dzms
= dz cos a == dzIs z z =::} -£llms + VIIs = dz sin a + dz cos a == dz
vI
The rotation
eis not affected by this transformation and thus
Therefore the three degrees of freedom at node 1 are transfon:ned as follows:
267
268
TRUSSES, BEAMS,AND FRAMES
The same relationship holds for the degrees of freedom at node 2. Thus the transformation between the global and the local degrees of freedom can be written as follows:
d] d2 d3 d4 ds d6
L=
Is -», 0 0 0 0
~ (x2 -
x])2
ms Is 0 0 0 0
0 0 0 0 0 0 1 0 0 0 ms Is 0 -ms Is 0 0 0
+ (Y2
0 0 0 0 0 1
H] VI
8] H2
===> d, = Td
v2
82
1 = coso: = s
- YI)2;
X -x
_2_ _];
L
m = sin 0: = s
Y2 - YI
L
where 0: is the angle between the local s axis and the global x axis measured counterclockwise, (Xl' Yl) and (x2' Y2) are the coordinates of the two nodes at the element ends, and L is the length of the element. It is easy to verify that the inverse transformation (i.e., local to global transformation) is given simply by the transpose of the T matrix,
Using this transformation, the element equations in the local coordinate system can be related to those in the global coordinate system as follows:
...
"'
k, d, =1', ===> k,Td
.., r'
=1',
Multiplying both sides by TT, we get' TTk,Td
=TTl',
.I
Noting that TTl', is the transformation of applied loads from the local to the global coordinates, we have the following element equations in terms of global degrees of freedom and applied nodal loads in the global directions:
kd
=1'
where and
It is possible to carry out the matrix multiplications and write the element equations in global coordinates explicitly. However, the resulting matrix is quite large and is not written here. The Mathematica implementation given at the end of this section shows the explicit expressions. For manual computations it is much easier to write the element equations in local coordinates first and then carry out the matrix multiplications using the numerical data. . The assembly and solution for global nodal unknowns follow the standard procedure. For computing the element solution, the global degrees of freedom for each element are
PLANE FRAMES
first transformed into the local degrees of freedom by multiplying them by the T matrix for the element. These local degrees of freedom are then used with the axial deformation and beam interpolation functions to get complete displacement solutions. In terms of notation used in this section, the appropriate expressions are as follows: Global to local transformation:
Axial displacement: u(s)
«
=( -T
N= (_s-L L II
f)(~J;
f)
Transverse displacement:
v(s) =
3' U- I! +1 ? 3
(
Fixed-end solution for uniformly distributed load:
To get the exact solution for. the distributed load case, the fixed-end solution is added to those elements that are subjected to distributed load qt. Thus a finite element model of a frame needs only one element between any two joints or between locations of concentrated loads. Finally, the axial forces, bending moments, and shear forces are computed by appropriate differentiation:
= EA d:~s) ;
Axial force:
F(s)
Bending moment:
M(s) = El
Shear force:
V(s) =
d2v(s)
ds 2
. '
d~;S);
If desired, the stresses in the elements can be computed using expressions given with axial deformation and beam elements.
269
270
TRUSSES, BEAMS, AND FRAMES
P
T
L
M
1
I--L-l--L
Figure 4.36. Plane frame
Example 4.10 Determine displacements, bending moments, and shear forces in the plane frame shown in Figure 4.36. Draw free-body diagrams for each element clearly showing all element end forces and moments. Use the following numerical data:
M=20kN·m;
2
E = 210 OPa;
L= 1m;
P=10kN;
A = 4 X 10- m
2
;
1= 4 X 10-4 m"
The simplest possible model for the frame is to use three elements as shown in Figure 4.37. Each node has three degrees of freedom as indicated in the figure. Use kilonewton-meters for numerical computations. The computed displacements will be in meters and stresses in kilonewtons per square meter. Specified nodal loads: Node
dof
Value
2 4
v2
-10 -20
84
Equations for element 1: / E = 2.1 X 108; I = 0.0004; A = 0.04; qs = 0.; qt = O. Nodal coordinates: Element Node
Global Node Number
1 2 Length = 1; Direction cosines:
x
y
100 2 1 0 Is
= 1, Ins = 0
Figure 4.37. Three-element model for the plane frame
Bending moment, M = EI dZv(s)/dsz = -5.s Shear force, V(s) = dM/ds = -5. Solution for element 2: Axial displacement, u(s) = N;'
(~J = -1.78571 x 1O-6s -
Axial force, EA du(s)/ds = -15.
0.0000685516
PLANEFRAMES
Transverse displacement, v(s)
=N;, [;;] = 0.0000487103s -
0.0000297619s 2
. d6 Bending moment, M = EI d 2v(s)/ds2 = -5. Shear force, yes) = dM/ds = 0 Solution for element 3:
Axial
.
displace~ent, u(s) =N[ (~J = 0.0000189484
Axial force, EA du(s)/ds = 0 Total transverse displacement, v(s) = -0.0000297619s3
-
0.0000297619s 2
- 0.0000108135s + 0.0000703373 Bending moment, M = EI d 2v(s)/ds 2 Shear force, Yes) = dM/ds = -15. Forces at element ends:
= -15.s -
5.
x
y
Axial Force
Bending Moment
Shear Force
1
0 1
0 0
0 0
0 -5.
-5. -5.
2
1 1
0 -1
-5. -5.
0 0
3
1 2
-1 -1
-15. -15. -0 0
-5. -20.
-15. -15.
The element end forces are shown in Figure 4.38. It can easily be seen that each element is in equilibrium. The finite element solution is exact.
Figure 4.38. Free-body diagram for the plane frame
273
274
TRUSSES,BEAMS, AND FRAMES
T L/{2
L
--..j
1
Figure 4.39. Plane frame with distributed load
Figure 4.40. Two element model for the plane frame
Example 4.11 Determine displacements, bending moments, and shear forces in the plane frame shown in Figure 4.39. Draw free-body diagrams for each element clearly showing all element end forces and moments. Use the following numerical data: (l kip = 1000lb).
q = 1 kip/ft;
L
= 15 ft;
1= 1000 in4
To model the frame, we need only two elements: one for the inclined member and one for the horizontal as shown in Figure 4.40. Each node has three degrees of freedom as I indicated in the figure. Using kip-inches, the solution is as follows: Equations for element 1: E = 30000; I = 1000; A Nodal coordinates:
= 100; qs = 0.; qt = -0.0833333
Element Node . Global Node Number 1 2
1 2
Length 180.; Direction cosines: 1s = 0.707107, Element equations in local coordinates: 16666.7 0 0 -16666.7 0 0
0 61.7284 5555.56 0 -61.7284 5555.56
0 5555.56 666667. 0 -5555.56 333333.
-16666.7 0 0 16666.7 0 0
Ins
0 -61.7284 -5555.56 0 61.7284 -5555.56
y
x
O.
O.
127.279
127.279
= 0.707107 0 5555.56 333333. 0 -5555.56 666667.
d1 dz d3 d4
ds d6
O. -7.5 -225. O. -7.5 225.
275
PLANEFRAMES
Global to local transformation, 0.707107 -0.707107 0 T= 0 0 0
0.707107 0.707107 0 0 0 0
0 0 0 0 1 0 0 0.707107 0 -0.707107 0 0
0 0 0 0.707107 0.707107 0
0 0 0 0 0 1
Element equations in global coordinates: 8364.2 8302.47 -3928.37 -8364.2 -8302.47 -3928.37
Adding element equations into appropriate locations and adjusting for essential boundary conditions, we have .
][U2) 25030.9 X I02.47 3928.37 8302.47 8425.93 1627.18 v2 [ 3928.37 1627.18 1.33333 x 106 82
=
[5.3033) -5.3033 225.
Solving the final system of global equations, we get {u2 = 0.000601607,v2
= -0.00125474,82 = 0.000168509}
Solution for element 1:
E = 30000; I = 1000; A = 100; qs = 0.; qt = -0.0833333 Length = 180.; Direction cosines: 1s = 0.707107, Ins ="0.707107 Nodal values in global coordinates,
Bending moment, M = EI d 2v(s)/ds2 = 0.858707s - 105.368 Shear force, Yes) = dM/ds = 0.858707 Forces at element ends:
x
y
Axial Force
Bending Moment
Shear Force
1
0 127.279
0 127.279
-7.69721 -7.69721
-288.462 -105.368
8.51719 -6.48281
2
127.279 307.279
127.279 127.279
-10.0268 -10.0268
-105.368 49.1988
0.858707 0.858707
PLANEFRAMES
0.859 ;J49.2 105-( ~ t f - - - - - ,
.
QL=lV~ 8.x/ 6.48
~
~r
vI0
0.859
7 1 105 . .
288·C Figure 4.41. Free-body diagram of frame
The element end forces are shown in Figure 4.41. It can easily be seen that the equilibrium is satisfied in each element and hence this is the exact solution.
+ MathematicafMATLAB Implementation 4.4 on the Book Web Site: Analysis ofplane frames Example 4.12 Determine displacements, bending moments, and shear forces in the box frame reinforced with a spring as shown in Figure 4.42. Use the following numerical data (1 kip = 1000Ib):
L
= 15 ft;
I
= 1000 in";
E
= 30,000 kips/irr': k = 200 kips/in;
P = 100 kip
We can easily create a finite element model of the entire frame using four plane frame elements and one spring element. However, since there are no specified essential boundary conditions, the resulting finite element system of equations will be singular. We can get some boundary conditions bytaking advantage of symmetry and modeling the upper half, as shown in Figure 4.43. Note the spring constant and the applied loads are both halved in the model. The ends must have zero rotations and zero vertical displacements due to
"''',~-''!;0-- P
Figure 4.42. Box frame reiuforced with a spring
277
278
TRUSSES, BEAMS, AND FRAMES
Figure 4.43. Finite element model for the half of the box frame
symmetry. The ends are free to move in the horizontal direction. The system can still move as a rigid body in the horizontal direction, and therefore the finite element equations will still be singular. To get a solvable system of equations, we must assign zero horizontal displacement to at least one of the nodes. For this frame, once again because of symmetry, the top of the frame cannot move in the horizontal direction. Thus we impose the zero horizontal boundary condition at that point. Using kip-inches, the solution is as follows: Specified nodal loads: Node
Spring force = k(u 3 - u l ) = 200(0.184555- (-0.184555)) = 73.822 (Tension)
4.9 SPACE FRAMES A space frame element is an extension of the plane frame element to three dimensions. The element loading and cross-sectional properties are described in terms of a local r-s-t coordinate system, as shown in Figure 4.44. The local t axis runs along the centroidal axis of the element. The local rand s axes are the principal moment-of-inertia axes for the cross section; the local r axis is along the axis of the maximum moment of inertia and the local s axis is along the axis of the minimum moment of inertia. The element includes axial force effects and bending effects due to applied loads in the r-t and s-t planes. In addition, the
Figure 4.44. Space frame element
279
280
TRUSSES, BEAMS, AND FRAMES
element includes torsional effects due to the twisting moment (moments about the taxis). Within the small-displacement theory, all these effects can be assumed to be uncoupled. The finite element equations are thus just a combination of the equations treating these effects individually. -For a space frame element, just knowing the element end coordinates is not enough to establish all three local element coordinates. We need additional information about one of the local principal axes. The simplest method is the so-called three-node method. Nodes 1 and 2 are at the element end points. These two nodes define element length and degrees of freedom. For each element a third point must also be identified to define the local r-t plane. Any node in the model can be used as the third point as long as the three points define the local r-t plane for the element. Using the three nodes, the local element axes are established as follows: (i) The local t axis is first established by creating a vector from node 1 to 2.
(ii) Since the cross product of two vectors is a vector that is normal to the plane defined by the vectors, the local s axis is established by taking the cross product of a vector from node 1 to 2 with a vector from node 1 to 3.
(iii) Finally the local r axis is established by taking the cross product of unit vectors in the sand t directions. m 1/1
.II I' 111
,"lin
The decision to use the third node to define the r-t plane is obviously arbitrary. We could have used it to define the s-t plane as well. There is no standard convention. Different authors and computer programs use the third node to define either the major bending axis or the minor axis. Thus, when using a computer program, it is very important to find out exactly what convention is being used; otherwise the results obviously will be erroneous. Detailed expressions for these vectors are developed later in this section. The following notation is used to/describe material and cross-sectional properties:
E
Young's modulus Shear modulus Area of cross section J Torsional constant Ip Polar moment of inertia Is =Imin Moment of inertia of cross section about s axis (minimum principal inertia) t, = I mnx Moment of inertia of cross section about r axis (maximum principal inertia) Length of element L
G A
The torsional constant is sometimes also 'denoted by kT or It. For circular cross sections J = Imax + Imin = Ip ' the polar moment of inertia. For other shapes J must be computed using methods of elasticity theory. Formulas for few common shapes are given in Figure 4.45. More formulas for a large number of different cross-sectional shapes can be found in the handbook by W. C. Young and R. G. Budynas, Roark's Formulas for Stress and Strain, seventh edition, McGraw-Hill, New York, 2002.
SPACE FRAMES
1 16 b b4 J =-ab 3 (- - 3.36- (l - - -4 » 16 3 a 1Za
9 a4 64
For b =a:J=-
Figure 4.45. Formulas for torsion constant J
4.9.1
Element Equations in Local Coordinate System
Considering axial forces, bending in both planes, and twisting effects, each node of a space frame element has six degrees of freedom; three translations, and three rotations, as shown in Figure 4.46. The nodal displacements and applied forces are positive when these quantities act along the positive coordinate directions. For applied moments and rotations the positive directions are based on the right hand rule. When the thumb of your right hand is pointing toward the positive coordinate direction, the curl of your fingers defines the positive directions for applied moments and rotations in the right-hand TIlle. The nodal degrees of freedom in the local coordinate system are defined as follows: dl' dz• d3 d4• ds• d 6
Displacements at node 1 Rotations at node 1
s; d g• d9
Displacements at node 2
d lO• d u ' d 12
Rotations at node 2
dz
Figure 4.46. Local degrees of freedom for the space frame element
281
282
TRUSSES, BEAMS, AND FRAMES
The axial displacements are related to the two axial degrees of freedom d l and d7 as follows:
where L is the length of the element. The applied loading in the local s-t plane will cause the displacement v(t) in the local s direction and rotations == dv/dt about the r axis. From beam bending the displacement v(t) is related to nodal degrees of freedom d2 , d6 , ds' and d l 2 as follows:
e
v(t)
J2.)[~~]. s
.e.L+L
=( I
d
2
'
. dl2 This loading causes bending moment about the r axis and shear force normal to the t direction: V(t)
(
= dM, = E dt.
r
(ddt3V) 3
For the loading applied in the local r-t plane, the displacement w(t) is in the local r direction and rotations
;'
ow -at
The finite element shape functions for this situation are the same as those for usual beam bending except for the change in sign for the rotation terms. Thus the displacement w(t) is related to nodal degrees offreedom d3,ds' d9 , and d ll as follows:
~:
_( _ t.L+lJ (3 )) [ ]. d ' 9 d ll This loading causes the bending moment about the s axis and shear force normal to the r direction: V (t) r
3
s = E1 (d W ) =_dM dt s dt 3
The elements in a space frame generally are also subjected to twisting moments. The governing differential equation for a bar subjected to twist is a second-order differential equation similar to the one for the axial deformation problem. Assuming twisting moments are
SPACE FRAMES
283
-applied only at the element ends (no applied distributed twisting moment along the span), 'the problem can be described in terms of the following differential equation:
GJd21jJ = 0 dt 2 where ljJ(t) is the rotation of the section about its taxis, G is the shear modulus, and J is the torsional constant. From the similarity of this equation to the one for the axial deformation problem, it is easy to see that a linear solution in terms of nodal degrees of freedom d4 and d 10 is as follows:
The twisting moment per unit length is related to the angle of twist by the following equation:
Mr(t) = GJ~~ By combining axial deformations, bending in r-t and s-t planes, and effects due to twisting actions, the total strain energy in a frarne element can be written as follows:
U
2V)2 1 dt + 2'
1 t' (d = 2'I Jot'EA (dU)2 dt dt + 2' Jo ei, dt2
2W)2 I t'. (dljJ)2 dt + 2' J GJ 7ft dt
Jot'm, (ddt2
o
It is convenient to express this in matrix form as follows: du
u=~ rL(f!E. 2
Jo
dt
[
EA
o
0
0
tu,
0
o
EI~
0
o
00
0] o
dt 2
d
v
dr2
d2w
~j ~;
dt
dt
Using the different displacement interpolations given above, the required derivatives can be written as follows: du
o
dt
o o
d2 v
dr
2
-r1 BT=
0 0 0
0 12r
0 6
l!-IJ 0
0
0
0 12r
0
0 6
l! - IJ 0
0 1
-r
0 6r
4
L - IJ
0
1
0 6r
0
IJ - L '0 0
0
0
r 4
0
6
0 121
IJ - l ! 0
0
0
0 6
IJ
121
t:
r
0
0
0 6r
2
0
0
0
L - IJ
IJ - L 0
1
0
0
t.
2
61
284
TRUSSES, BEAMS, AND FRAMES
Thus the strain energy can be written as follows:
where 0
0 0
0 0
EI s
e-- [EA 0 st, 0 0
k=
LL
1]
0
BeBT dt
Carrying out matrix multiplications and integration, we get the following space frame stiffness matrix in its local coordinates: EA
IIIIIII 11111" t.. ,~~; U!h~
"It,,, II,·tll "1-'
.,." "II
". :11'
1,U,
'lIIit.
k=
0
T 0
12EI, L3
0 0 '0
0 0 0
0 EA
-T 0
0 0 0 0
6El,
IF 0
_12EI, L3
0 0 0
~ L
0 0
0 0 0
11£/, L3
GJ
0
T 0
-~ L2 0 0 0
0 0 0 0 GJ -T 0 0
12EI,
-7 0 -~ L2 0
0 0
EA
0 ~ 2 L
6El,
-IF
0 0 0 0
0 0 0
0 ~ L
4El, L
0 0 0
EA
0
6El, L2
0 2E1,
L 0
-T 0
-~ L2 0 ./ 0 0 2EI,
L
T 0
0 0 0 0
0 _12EI, L3
0 0 0 -~ L2 0
0 0
0 0 0 GJ -T 0 0 0 0 0
_12EI, L3
0 ~ 2 L
0 0 0
12El, L3
12~/, L
0 0 0
-~ L2 0 2El
-L '
0 0 0 ~ 2
6EI, L2
0
0 ~ 2 L
0 0 0 3§.!L L
0
-~ L
T 0
~
0 0 0'
0
0
~
L
GJ
0
-~ L2
0 0
0 L
L
In a similar manner, the equivalent nodal load vector can be established by considering the work done by the distributed applied forces along the local coordinate directions qr and qs as follows:
Writing the displacement expressions in a matrix form, we have 0 ( vet) ) = [ wet) 0 =NTd
2t3 L3
-
3t2 L2
0
+1
0 2rJ
TJ -
0 2
3t L2
+
0
1 0 -lJ + T - t t3
2t2
t3
lJ
2t
2
L
0
+t
J[ ~] d 12
SPACE FRAMES
Using these, the work done can be written as follows:
where rq is the equivalent load vector. If all distributed forces are assumed constant over an element, then the integration can easily be carried out to give the following equivalent load vector: rT q
=
{o
Z Z z qsL q,L 0 _ q,Lz qsL 0 qsL qrL 0 q,.L _ q,L } '2 ' 2 " 12' 12' , 2' 2' , 12' 12
4.9.2 local-to-Global Transformation The stiffness matrix and the equivalent load vector derived so far have been in terms of a local coordinate system. Since different elements in a space frame will generally have different local axes, before assembly these matrices must be transformed to a global coordinate system that is common to all elements in the frame. In the global x, y, and z coordinate system the nodal displacements are identified as follows:
v]> wI ex p ey]> ezl Ll1,
LiZ'
vz' Wz
exz' eyZ' ezz
x, y, and z displacements at node 1 Rotations about x, y, z axes at node 1 x, y, and z displacements at node 2 Rotations about x, y, z axes at node 2
The corresponding nodal forces and moments in the global x, y, and z coordinate system are as follows: Fxl' F;,I' Ftl M.d'
lvIY1'
»:
Ftz
F.tZ' FyZ' lvIxz ' lvIyZ' lvIa
Applied forces in the global x, y, and z at node 1 Applied moments about the global x, y, and z at node 1 Applied forces in the global x, y, and z at node 2 Applied moments about the global x, y, and z at node 2
The local-to-global transformation matrix is developed by considering three components of displacements and rotations at each node as vector quantities. Thus the complete transformation matrix is a 12 x 12 matrix consisting offour identical 3 x 3 rotation matrices as follows:
where H is a 3 x 3 three-dimensional rotation matrix and 0 is a 3 x 3 zero matrix. The rotation matrix H transforms a vector quantity from the local to the global coordinate system.
285
286
TRUSSES, BEAMS,AND FRAMES
Within the small-displacement assumption, the nodal displacements, rotations, forces, and moments are all vector quantities and can all be transformed using this H matrix. The components of a vector along the local s, t, and r coordinates are simply the sum of projections of its x, Y, and z components along the local axes. In matrix form the transformation can be written as follows:
where 1( is the cosine of the angle between the t and x axes (direction cosine). The other terms have a similar meaning. Thus nine direction cosines are needed to establish the H matrix for each element in a space frame. Explicit expressions for these direction cosines are given later in the section. Using this transformation matrix, the element equations in the local coordinate system can be related to those in the global coordinate system as follows:
Multiplying both sides by TT, we get
r,
Noting that TT is the transformation of applied loads from the local to the global coordinates, we have the following element equations in terms of global degrees of freedom and applied nodal loads in the global directions: I
kd =r
where and
Three-Node Method for Calculating Direction Cosines Just knowing the element end coordinates is not enough to establish all nine direction cosines in the rotation matrix H. We need additional information about one of the local principal axes. The simplest method is the so-called three-node method. Nodes I and 2 are at the element end points, as in the case of a plane frame element. A third node is used to define either the local r-t or the s-t plane. In the following development it is assumed that the third node is used to define the local r-t plane, as shown in Figure 4.44. It is important to note that the third node is used for defining the element orientation alone. It does not determine the element length or any other parameter used in the element equations. In a given frame any suitable node can be used to define the third node as long as this node, together with element nodes I and 2, determines the r-t local plane for the element. The third node cannot be collinear with element nodes I and 2 because in this case the three nodes do not form a plane.
SPACE FRAMES
The coordinates of the three nodes for an element are denoted by (xl' Yl' z.), (X2• Y2' Z2)' and (x3• Y3' Z3)' Using these COOrdinates, the nine direction cosines can be computed as follows. The local t axis is defined first by a vector V12 pointing from node 1 to node 2 of the element: .
where i,j, k are unit vectors along the global coordinate directions. The length of the vector V12 is the element length L = ~ (x2 - x 1)2 + (Y2 local t axis is therefore given by .
Yl)2
+ (Z2 -
Zl)2.
A unit vector along the
The three direction cosines defining the local t axis are thus
1
= X2 -xl.
= Y2 -Yl.
m
L'
I
L'
I
111
Z -Z = _2_ _ 1
L
The local s axis is normal to the plane defined by nodes 1, 2, and 3. A vector along the local s axis is obtained by taking the cross product of the vector V12 with vector V13 that goes from node 1 to node 3. Thus we have
i
Vs = Vl3 X V12 = det x3 [
Y3 - YI
X2-XI
Y2-YI
Xs = Y3(Z2 -
Zl)
+ Y2(~1
Y, = X3 (z, -
Z2)
+ Xl (Z2 -
Z,
= X3(Y2 -
j xl
- Z3)
Z3)
Yl)'± X2(YI - Y3)
+ YI (Z3 -
Z2)
+ X2(Z3 -
Zl)
+ Xl (Y3 -
Y2)
The length of this vector Vs is
Therefore a unit vector along the local s axis is given by
The three direction cosines defining the local s axis are thus y Ls '
=~.
nt S
287
288
TRUSSES, BEAMS,AND FRAMES
Finally, the local r axis is normal to the s-t plane. It can be defined by taking the cross product of the unit vector along the s axis with the one along the t axis
where
Example 4.13 An f-shaped frame element is oriented in such a way that its longitudinal axis is along the global z axis and its minor moment of inertia axis makes an angle of 30° with the global x axis. The element is 100 em long. Establish the 3 X 3 rotation matrix H for the element. Assuming node 1 is at (0, 0, 0), node 2 should be at (0, 0, 100). According to the convention adopted here, the third node should define the major moment of inertia axis. The angle between the global x axis and the major moment of inertia axis will be 60°. Thus, to define the orientation of this element, we choose a third point located at 100(- cos(60), sin(60), 0). This will establish the t - s - r coordinate system for the element, as shown in Figure 4.47. The computations of direction cosines using the three-node method are as follows: Nodal coordinates:
1
x
y
z
0
o o
0
2 0 3 /-50
100.
50·..j3· 0
Vector from node 1 to 2, VIZ = (O, 0,100.) Element length, L = 100. z,t
Figure 4.47. Local and global axes for a space frame element
SPACE FRAMES
" Unit vector,
VI = vlzlL = (O, 0,1.)
. Vector from node I to 3, Vl3 ={- 50, 50{3, 0) Vector, Vs = VI3 X VIZ = (8660.25, 5000., 0) Length, L, = 10000. Unit vector, Vs = V/L s = (O. 866025,0.5, 0) Unit vector, Vr = VI X Vs = (-0.5,0.866025,0.) Thus the rotation matrix is as follows:
H
=[
~.866025 0.866025 ~.5 ~'J O.
-0.5
4.9.3 Element Solution For computing element solutions, the global degrees of freedom for each element are first 'transformed into the local degrees of freedom by multiplying them by the T matrix for the element. These local degrees of freedom are then used with the axial deformation and beam interpolation functions to get complete displacement solutions. Finally the axial forces, bending moments, and shear forces are computed by appropriate differentiation. In terms of notation used in this section the appropriate expressions are as follows: Global to local transformation:
Axial displacement:
I)(~~);
i) Axial force: F(t)
=EA duCt) . dt '
Transverse displacement in the local s direction:
O:5.t:5.L
289
290
TRUSSES,BEAMS.AND FRAMES
Transverse displacement in the local r direction:
wet) =( 1 -
3(2
L2
+
2(3
L3
-(t _
2(2
L
+ ..t.) 2
O.s ts L
L
Bending moments:
Mr(t)
= EIr (~:~ ) ;
Ms(t)
= -EIs ( ~:n;
0 :5, t :5, L 0:5, t :5, L
Shear forces: III
iii
Vs(t) =
'"
~I
'~I
V,(t)
d~r =EIr ( ~:~ ) ;
0 :5, t :5, L
=- d;s =EIs ( ~:t;) ;
0 :5, t :5, L
Twisting moment: !
if!(t) = ( 1 -
i:
4
Lt ) ( ddlO).'
dif! M(t) = GJ dt Example 4.14 Analyze the one-story three-dimensional frame shown in Figure 4.48. The height of the columns is 12ft and the length of the beams is 10 ft. Each beam is subjected to a uniformly distributed load of 2 kip/ft in the downward direction. I-shape sections are used for both columns and beams with the arrangement shown in the figure. The columns are connected to the foundation through simple connections that do not resist moments. The material is steel with E = 29000kip/irr' and G = 11200kip/irr', The section properties are as follows: Beams:
J=43in 4 ;
I max = I, = 450 in";
Columns:
J = 60in4 ;
I max
=I, = 650 in";
=Is =32 in2 I min =Is = 54 in 2 I min
SPACE FRAMES
3 2
2
3
y
x Flgure 4.48. One-story space frame
Taking advantage of symmetry, we model a quarter of the frame using three elements. Because of symmetry, the boundary conditions at nodes 3 and 4 are as follows: Node 3:
£1=0;
Node 4:
v=O;
ex = 0;
The distributed load is applied to the elements in their local coordinates. Therefore, to assign proper direction and sign to the distributed loads, we must carefully establish the local coordinates for the elements as follows: Element 1: Nodes 1,2, and 4
=> t axis along global z; s axis along global x; r axis along global y Element 2: Nodes 2, 3, and 4 ,
=> t axis along global x; s axis along global -z; r axis along global y Distributed ~oad: q, ":' 0; q,
= fz kip/in
Element 3: Nodes 2, 4, and 3
=> t axis along global y; s axis along global z; r axis along global x Distributed load:' qr = 0; qs = -
fz kip/in
Matrices for this example are too large for printing. A complete solution can seen on the book website. Solving the final system of global equations, we get
dll Bending moment, M, = -E1, d 2w(t)/dt 2 Shear force, Vr = -dM/ dt = 0.180999
= -0. 180999t
FRAMES IN MULTISTORY BUILDINGS
Solutions over the remaining elements can be obtained in a similar manner:
Forces and Moments at Element Ends x
y
Z
Axial Force
V.
Vr
'M r
Ms
Mt
0 0
0 0
0 144.
-20.
0.889955 0.889955
0.180999 0.180999
0 128.154
0 -26.0639
0 0
2
0 60.
0 0
144. 144.
-0.889955 -0.889955
0 0
128.154 -171.846
0 0
0 0
3
0 0
0 60.
144. 144.
-0.180999 -0.180999
-10. 0 10. 0
0 0
-26.0639 273.936
0 0
0 0
~
~20.
MathematicafMATLAB Implementation 4.5 on the Book Web Site:
Analysis of space frames
4.10
FRAMES IN MULTISTORY BUILDINGS
From a structural point of view most modem multistory buildings consist of a grid of structural frames. The floor system, consisting of metal decks and concrete slabs, span across the frames and tie everything together to create a three-dimensional structure, as shown in Figure 4.49. Because of large dimensions, the floor system within its own plane is essentially rigid and is known as a rigid diaphragm. Thus individual frames in a building do not behave as their isolated counterparts. We must analyze the entire three-dimensional system. With the assumption of an in-plane rigid-floor system, each story has only three independent in-plane degrees of freedom: the two displacements in the x and y directions (£I and
Figure 4.49. Multistory building
293
294
TRUSSES, BEAMS, AND FRAMES
Figure 4.50. Rigid zone at beam-column connections
v) and a rotation about the z axis (8). Thus, to analyze a building, we use standard frame element equations and assemble contributions from all frames in the building in the usual manner. For each story we define constraints that relate all in-plane degrees of freedom to one node that we call the master node for that story. Identifying the master node degrees of freedom by subscript m, the constraints for degrees of freedom at any other node in the story are expressed as follows: l '1.
'Ii,
i
= 1,2, ...
where Xi and Yi are the coordinates of node i. Joints in a building frame also require special modeling care. The column dimensions in a typical building are of the order 14 to 20 in while the beams may vary in depth from 21 in to 30 in. When creating frame models using centerline dimensions, the joint zone with these large member sizes is fairly large. Very little deformations are expected within this jointzone. Thus rigid joint zones are typically created when analyzing building frames, as shown in Figure 4.50. The nodes are placed outside of the joint region. The following constraints are defined between the degrees of freedom at these nodes to create the effect of the rigid joint zone:
After incorporating these constraints, the rest of the analysis follows the standard procedures used in other examples in this c h a p t e r . -
PROBLEMS
PROBLEMS Plane Trusses
4.1
Determine joint displacements and axial forces in the three-bar pin-jointed structure shown in Figure 4.51. All members have the same cross-sectional area and are of the sarne material, A = 1 in2 and E = 30 X 106 lb/irr'. The load P = 15,000 lb. The dimensions in inches are shown in the figure. 72
o • )I--------------+~-p (in)
o
108 Figure4.51.
4.2 Determine joint displacements and axial forces in the truss shown in Figure 4.51 if the support where the load is applied settles down by ~ in. 4.3 Taking advantage of symmetry, determine joint displacements and axial forces in the three-bar truss shown iIi Figure 4.52. All members have the same cross-sectional area and are of the same material, A = 0.001 m2 and E = 2000Pa. The load P = 20kN. The dimensions in meters are shown in the figure. One of the goals for this problem is for you to learn the correct use of symmetry. Thus no credit will be given if you do not-use symmetry even'if your solution is correct. 3
o -4
o Figure4.52.
(m)
4
295
296
TRUSSES, BEAMS, AND FRAMES
4.4 Determine joint displacements and axial forces in the truss shown in Figure 4.53. All members have the same cross-sectional area and are of the same material, A = 2 in2 and E = 30 x 106lb/in2 . The load P = 30,000 lb. The dimensions in feet are shown in the figure. 6
p
30
o (ft)
o
8 Figure 4.53.
4.5
Determine joint displacements and axial forces in the truss shown in Figure 4.53 if in addition to the applied force the lower left support moves horizontally toward the right by ~ in.
4.6
Determine joint displacements and axial forces in the truss shown in Figure 4.54. Note the diagonals are not connected to each other at their crossing. The crosssectional area of vertical and horizontal members is 30 x 10- 4 m2 and that for the diagonals is 10 x 10- 4 m2 . All members are made of steel with E = 210 GPa. The load P = 20 kN. The dimensions.in meters are shown in the figure. 6
(}-------.,----{)--- p
o (m)
o
8 Figure 4.54.
Space Trusses
4.7
Determine joint displacements and axial forces in the space truss shown in Figure 4.55. The cross-sectional area of members is 15 x 10- 4 m2 . All members are made
PROBLEMS
p
3P
Figure 4.55.
of steel with E = 210 GPa. The load P = 10 leN is applied in the -z direction and 3P in the x direction. The nodal coordinates in meters are as follows:
1 2 3 4 4.8
x
y
z
O. O. 6.9282 -6.9282
O. 8. -4. -4.
10. O. O. O.
Determine joint displacements and axial forces in the space truss shown in Figure 4.56. The cross-sectional area of members is 15 x 10- 4 m2 . All members are made
p Figure 4.56.
297
298
TRUSSES, BEAMS, AND FRAMES
of aluminum with E = 70 GPa. The load P = 10 kN is applied in the The nodal coordinates in meters are as follows:
1 2 3 4 5
x
y
Z
O. -3. -3. 3. O.
4. 2. O. O. O.
O. 5. O. O. 5.
-z direction.
Thermal and Initial Strains
4.9
A copper rod 1.4 in in diameter is placed in an aluminum sleeve with inside diameter 1.42 in and wall thickness 0.2 in, as shown in Figure 4.57. The rod is 0.005 in longer than the sleeve. A load P = 60,000 lb is applied to the assembly through a large bearing plate that can be considered rigid. Determine stresses in the rod and the sleeve. The modulus of elasticity for copper is 17 x 106 psi and that for aluminum is 10 X 106 psi. Rigid plate
II~
p
11
'"
II.II,11; Gap=0.005
"
Figure 4.57.
4.10
Determine joint displacements and axial forces in the truss shown in Figure 4.51 if instead of load P the diagonal member experiences a temperature decrease of 70"F. The coefficient of thermal expansion is a = 6 x 10-6j"F.
4.11
Determine joint displacements and axial forces in the truss shown in Figure 4.53. During construction the diagonal member was fabricated in too short and was forced to fit into the assembly through heat treatment.
4.12
Determine joint displacements and axial forces in the space truss shown in Figure 4.56 if instead of load P the member between nodes 1 lind 2 experiences a temperature rise of 50"C. The coefficient of thermal expansion is a = 23 x 1O- 6j"C.
!
Spring Elements
4.13
Determine support reactions and forces in the springs for the assembly shown in Figure 4.58.
Figure 4.58.
PROBLEMS
4;14
Determine support reactions and forces in the springs when the assembly is forced to close the gap g, as shown in the Figure 4.59.
Gap,g Figure 4.59.
Beam Bending 4.15
A fixed-end beam is subjected to a concentrated load at the midspan P.= 200 Ib, as shown in Figure 4.60. The beam has a rectangular cross section with width = 12 in ' and height = 1 in. The length of the beam is L = 200 in and its modulus of elasticity is E = 107 lb/in'', Taking advantage of symmetry, use only one beam element to determine the maximum bending and shear stresses in the beam. p
,. Figure 4.60. Fixed-end beam
4.16
Two cantilever beams are joined together through a simple pin, as shown in Figure 4.61. The concentrated load at the midspan is P = 200 lb. The beam has a rectangular cross section with width = 12 in and height = 1 in. The length of the beam is L = 200 in and its modulus of elasticity is E = 107 lb/irr'. Taking advantage of symmetry, use only one beam element to determine maximum bending and shear stresses in the beam. p
I"
--~I>I""f----
L/2
------j
Figure 4.61. Beams connected through' a pin
299
300
TRUSSES, BEAMS, AND FRAMES
4.17
Tho cantilever beams are joined together through a simple pin, as shown in Figure 4.62. The E1 = 107 for the right beam and is twice that for the left beam. The concentrated load at the hinge location is P = 200 lb. Determine the displacement and bending moment distribution over the beams. p
- - L/4 ~;-------
"I
3L/4
Figure 4.62. Beams connected through a pin
4.18
A two-span beam is subjected to a moment, as shown in the Figure 4.63. Find the resulting displacement, shear force, and bending moments. Compute shear and bending stresses at the middle support in the beam. Use the following numerical data (1 k = 1000 lb). .
E = 29,000 k/in 2 ;
L= 180 in;
M= 1000k·in
Beam cross section: Standard l-shape (W16 x 26) with 1 = 301 in" and dimensions as shown in the figure.
I-
I
+
L h
L
:;~~oo;~J~
= 15.69
br =5.5
I
-I
-be-'
Figure4.63.
4.19
Immediately after construction the right support of the beam shown in Figure 4.64 undergoes a settlement equal to A. Find the resulting displacement, shear force, and bending moments. Compute shear and bending stresses at the middle support in the beam. Use the following numerical data:
L=8m;
E
=200 GPa;
A= lOmm
Beam cross section: Standard I-shape with 1 = 125.3 X 10- 6 m" and dimensions (in min) as shown in the figure.
PROBLEMS
r
+
L
h br
= 399 = 140
tw
~
L
= 6.3
tr = 8.8
IT1 .... tr
......b r
Figure 4.64.
4.20
Find displacements and draw shear force and bending moment diagrams for the beam shown in Figure 4.65. Assume E = 210 GPa, I = 4 X 10-4 m", L = 2 m, P = 10 leN, M = 20 leN· m.
Figure 4.65. Beam with variable section
4.21
Two uniform beams are connected together through a spring as shown in Figure 4.66. Determine deflections, bending moment, and shear forces in the beams. What is the force in the spring? Tile numerical values are
E
= 104ksi;
L
= 100in;
P
= 100 kips;
Use kip . in units in calculations.
k
/jco--
L
-+-
L
-1
Figure 4.66. Two beams connected througha spring
k
= 2000 kip/in
301
302
TRUSSES, BE~MS, AND FRAMES
4.22
Consider a uniform beam simply supported at the ends and spring supported in the middle, as shown in Figure 4.67. Taking advantage of symmetry, analyze the beam using only one element. Assume the following numerical data:
E
L=4m;
= 70 GPa;
I
= 2 X 104 rnm";
k= 400N/mm;
P
= 20kN
p
.&~ =-.'EI .- ~·t-o-~~o,c,% k EI I---
·1
L
L---j
Figure 4.67. Beam supported by a spring
4.23
The left half of a beam is subjected to a uniformly distributed load q = 1 lb/in, as shown in Figure 4.68. The beam has a rectangular cross section with width = 12 in and height = 1in. The length of the beam is L = 200 in and its modulus of elasticity is E = 1071b/in2 • Create the simplest possible finite element model for this beam. Using this model, determine the bending and shear stresses in the beam at U 4 from the left end. q
Using only one element, find the midspan displacement, bending moment, and shear force for the beam shown in Figure 4.69. Note that the loading is large enough to cause the gap to close. Use the following data: E = 104 ksi,I = 1000 in", L = 100 in, q = 1 k/in, gap = 0.25 in. q
Gap'A~
I·
L
Figure 4.69. Single span beam with a gap
~I
PROBLEMS
4.25 Compute stresses in the two-span beam shown in Figure 4.70. There is a small gap between the middle support and the beam: L
=6m;
q
= lOkN/m;
= 200 GPa;
E
Gap D. = 5 mm
Beam cross section: rectangular with width = 120 mID and height = 300 mrn.
Figure 4.70. Two-spanbeam
= 27 Ib/in over the right span and a moment M = ~ lb . in at the free end, as shown in Figure 4.71. The beam has a rectangular cross section with width = 12in and height = 1in, giving A = 12 in 2 and 1= 1 in", The length of the beam is L = 1 in and its modulus of elasticity is E = 11b/in2 • (a) Create the simplest possible finite element model for this beam. Write element equations, assemble to form global equations, and obtain the reduced system of equations, taking boundary conditions into consideration. (b) After a analysis of this beam it is found that the rotation at the free end is -&. rad and that at the leftsupport is --&. rad. Determine the vertical displacement and bending moment at point a in the beam.
4.26 A beam with an overhang is subjected to a uniformly distributed load q
q
1""'1--- L/3 - - - l l > l < o I - - - L/3
- - - ..i1!>1<>l - * - - - L/3
---!13>l
Figure 4.71.
4.27 Using the simplest possible finite element model, determine the maximum bending and shear stress in a uniform beam simply supported at the left end and spring supported at the right end, as shown in Figure 4.72. Assume the following numerical data:
a
= 5 kN/m;
L = 5 m;
E = 200 GPa;
303
304
TRUSSES, BEAMS, AND FRAMES
y q(x) = a xfL
EI
I-- L/3· -I.. . - -- 2L/3 --~~I Figure 4.72. Uniform beam subjected to a linearly increasing load
4.28
Two overhanging beams are joined together through a simple pin connection as shown in Figure 4.73. Determine displacement, shear force, and bending moments. Use the following numerical data:
L= 8m;
E
=200 GPa;
q =3kN/m
Beam cross section: Standard l-shape with 1= 125.3 X 10- 6 m" and dimensions (in rom) as shown in the figure: q
A··-··-·"_· I----
~WLUjl.;Z£;~ . ~.. '=Z>.
L ---'-Ooo,..I.t--.7L--!-:3L.-tI.. - - L ---<>j
h =399 bf;;' 140
tw tf
=6.3
= 8.8
Figure 4.73.
Plane Frames
4.29
A 300-rom-wide and 100-rom-thick bar is supported and loaded as shown in Figure 4.74. Determine displacement, shear force, and bending moments. Assume E = 10 GPa.
PROBLEMS
4
8 3
o - - - - - - - - - (m)
o
2
3
Figure 4.74.
4.30
Determine displacements, bending moments, and shear forces in the plane frame shown in Figure 4.75. Draw free-body diagrams for each element clearly showing all element end forces and moments. Assume q = 10 kNlm, L = 2 m, E = 210 GPa, A = 4 X 10- 2 m2 , and I = 4 X 10- 4 m". Take advantage of symmetry.
q
I 1 L
Figure4.75.
4.31
Determine displacements, bending moments, arid shear forces in the plane frame shown in Figure 4.76. Draw free-body diagrams for each element clearly showing all element end forces and momentS. Assume q = 10 kN(m, L = 2 m, E = 210 GPa, A = 4 X 10- 2 m2 , and I = 4 X 10- 4 m". Take advantage of symmetry. Use kN . m units.
305
306
TRUSSES, BEAMS,AND FRAMES
I L
t L
1 Figure 4.76.
4.32 The element stiffness matrices for vertical and horizontal elements of a plane frame have already been computed and assembled into a 9 x 9 global matrix as follows:
It is decided to brace the frame by an inclined truss member as shown in Figure 4.77. Write down the stiffness matrix for the inclined truss member. Assemble it into the global matrix. Determine nodal displacements due to a downward concentrated load
y
P
5 4
3
2
3 2
4
0 0
2
3
Figure 4.77.
4
5
x
PROBLEMS
of P = 50 kN applied at the corner. For all members, E and A = 10-2 m2 .
4.33
= 210 GPa, I = 10-4 m",
Determine displacement, shear force, and bending moments in the plane frarne shown in Figure 4.78. Use the following numerical data: P = l2kN;
E
= 200 GPa;
Cross-section: Standard l-shape with A
q
= 3lcN/m
= 4.95 X 10-3 m2 and I = 125.3 X 10-6 m",
p
6
3
o (m)
o
6
9
Figure 4.78.
4.34
Two 200 mm x 200 mm square bars are joined through a pin and loaded as shown in Figure 4.79. Determine displacement, shear force, and bending moments. Assume E = 10 GPa and P = 20lcN e .
6
3 1.5
o o
(m)
4 Figure 4.79.
6
8
307
308
TRUSSES, BEAMS,AND FRAMES
4.35
The lower portion of an aluminum step bracket, shown in Figure 4.80, is subjected to a uniform pressure of 20 Nzrnnr'. The left end is fixed and the right end is free. The dimensions of the bracket are thickness = 3 mm, L = 150 mm, b = 10 mm, and h = 24 mm. The material properties are E = 70,000 Nzrnm? and v = 0.3. For a preliminary analysis determine stresses in the bracket using a simple model consisting of three plane frame elements.
\
L
Free \ f--
h/2
-f--
h/2---\
Figure 4.80. Step bracket
4.36
The top surface of an S-shaped aluminum block, shown in Figure 4.81, is subjected to a uniform pressure of 20 N/mm2 . The bottom is fixed. The dimensions of the block are t = 3 mm, L = 15 mm, b = 10 mm, and h = 24 mm. The material properties are E = 70,000 N/mm 2 and v = 0.3. For a preliminary analysis determine stresses in the block using a simple model consisting of four plane frame elements.
/
Figure 4.81. S-shaped aluminum block
4.37
Figure 4.82. Concrete dam
The cross section of a concrete dam is shown in Figure 4.82. Using a simple plane frame model, estimate stresses in the-darn due to self-weight of the dam and the water pressure. Use a reasonable number of elements and assign average section properties to each element. Assume unit thickness perpendicular to the plane. The depth of the water behind the dam is 18 m, The density of concrete is 2400 kg/m'' and that of water is 1000 kg/m". The modulus of elasticity of concrete is 30 GPa.
PROBLEMS
4.38
The side view of a pry bar is shown in Figure 4.83. The cross section of the bar is rectangular with thickriess = ~ in and width (perpendicular to the plane of paper) = 1.2 in. The other dimensions (in inches) are shown in the figure. The material properties are E = 29 X 106 psi and v = 0.3. A load of P = 200 lb is applied at the center of the handle. Using a suitable number of plane frame elements, determine maximum deflection and stress in the bar. p
Figure 4.83. Pry bar
Space Frames 4.39
A cantilever beam is supported by a beam fixed at both ends, as shown in Figure 4.84. Analyze the system when it is subjected to loading as shown. Use the following numerical data:
= 1 m;
L=4m;
b
Beams, W360
x 44.6:
= 100N/m; E = 200 GPa; G = 100GPa A =5.71 X 10-3 m2 ; J = 0.1729 X 10- 6 m";
q
Imax = 121.1 X 10-6 m"; Beams, W310
x 38.7:
A = 4.94 I max = 84.9
X
X
10- 3 m 2 ; 6
10- m";
Figure 4.84.
I min = 8.16 X 10-6 rn" J = 0.1421 X 10- 6 m"; l min = 7.2 X 10-6 m"
309
310
TRUSSES, BEAMS, AND FRAMES
4.40
Two beam cantilever out of an Hsframe, as shown in Figure 4.85. The columns are fixed at the top and the bottom. Analyze the frame when it is subjected to loading as shown. Use the following numerical data. L = 20 ft;
h = 15 ft;
Beams, W16 x 40:
Columns, W12 x 72:
b = 20 ft;
P = 30 kips;
=0.851 in"; l min =28.9 in"
= 21.1 in2 ;
J = 3.062 in":
A
A
Imax =597 in"; E
= 29,000 ksi;
q = 2kips/ft;
= 11.8 in ; Imax =518 in"; 2
G = 11,200 ksi
/ Figure 4.85.
J
I min = 195 in"
w = 0.5kip/ft
CHAPTER FIVE hi
-&
"dHll
. , PFHH
s
iH
FW
;a
54'
TWO-DIMENS~ONALELEMENTS
In the previous chapters the basic finite element concepts were demonstrated through simple one-dimensional elements. For most of those problems either the exact solution is known or there are several other numerical methods available that may be more convenient to use than the finite element method. This is not the case for two- and three-dimensional problems. The exact solutions are rare for these problems. Most other numerical methods are not as widely applicable as the finite element method. One of the key advantages of the finite element method is that the fundamental concepts extend easily from onedimensional to higher dimensional problems. The only complication is that the required integration and differentiation become more difficult, but these are calculus-related problems and have nothing to do with the finite element concepts. A review of vector calculus, especially integration over two- and three-dimensional domains, will make the transition from one-dimensional problems to two- and three-dimensional problems a lot smoother. In this chapter the basic finite element concepts are illustrated with reference to the following partial differential equation defined over an arbitrary two-dimensional region, such as the one shown in Figure 5.1: -a (au) k + -a ( k -au) + pu + q ax .tax ay Yay
=0
where kxCx, y), kyCx, y), p(x, y), and q(x,y) are known functions defined over the area. The unlrnownsolution is u(x, y). The equation can easily be recognized as a generalization of the one-dimensional boundary value problem considered in Chapter 3. Steady-state heat flow, variety of fluid flow, and torsion of planar sections are some of the common engineering applications that are governed by the differential equations which are special cases of this general boundary value problem. 311
--
312
TWO-DIMENSIONAL ELEMENTS
y
Normal, n
x
Figure 5.1. Two-dimensional solution domain
The differential equation is second order and therefore, based on the discussion in Chap.ter 2, the boundary conditions involving u are essential and those involving its derivatives are natural boundary conditions. The area of the solution domain is denoted by A and its boundary by C. The boundary is defined in terms of a coordinate e that runs along the boundary and an outer unit normal to the boundary n, as shown in Figure 5.1. The x and y components of the normal vector are denoted by nx and l1y:
The formulation presented in this chapter can handle boundary conditions of the following types: (i) Essential boundary conditioy(~pecified on portion of the boundary indicated by Ce:
uCe) specified on C, where e is a coordinate along the boundary. (ii) Natural boundary condition specified on portion of the boundary indicated by Cn :
au au lex ax I1x + ley ay ny = aCc)u(e) + f3(c) on CIl where aCe) andf3(c) are specified parameters along the boundary. When lex = ley == k, the left-hand side is the derivative of u in the direction of the outer normal to the boundary, and the boundary condition is expressed as
au au) au k ( ax nx + ay ny == k an = au + f3 Observe carefully that this expression does not allow specification of any arbitrary first derivative of u. Only the normal derivative can be specified. For example, on a boundary
SELECTED APPLICATIONS OF THE 2D BVP
that is perpendicular to the x axis, the components of the unit normal to the boundary will ben, = 1 and ny = 0 and thus' only au/ax can be specified by this form. The reason for choosing the normal derivative is because it gives rise to a convenient weak form. Furthermore, for most physical problems it is natural to specify the normal derivative or it is possible to use mathematical manipulations to express the derivative in this form. Thus it is not a severe limitation for practical problems. A few selected applications that are governed by the differential equation of this form are presented in the first section. The second section presents a general finite element formulation for the problem. Rectangular and triangular elements for solution of this differential equation are presented in the last two sections in this chapter. These elements are used to obtain approximate solutions for a variety of problems.
5.1 5.1.1
SELECTED APPLICATIONS OF THE 2D BVP Two-Dimensional Potential Flow
The problem of two-dimensional frictionless and incompressible fluid flow, lmown as potential flow, is governed by the following two equations: 1. Continuity equation: ~
+~ =0
2. Irrotational flow condition:
t; - ~ = 0
where u(x, y) and v(x, y) are the fluid velocities in the x and y directions. Two different formulations are used to find solutions of these equations and are known as stream function and potential function formulations: In the stream function formulation it is assumed that a scalar function, called the stream function ifJ(x, y), exists that is related to the fluid velocities in the x and y directions as follows:
aifJ
u=-;
ay
aifJ
v=-ax
With this assumption the continuity equation is satisfied exactly while the irrotational flow condition yields the following second-order partial differential equation:
In the potentialfunction formulation it is assumed that a scalar function, called the potential function ¢(x, y), exists that is related to the fluid velocities in the x and y directions as follows:
a¢ ay
v=-
313
314
TWO·DIMENSIONAL ELEMENTS
y
x
Figure 5.2. Flow around an object
With this assumption the irrotational flow condition is satisfied exactly while the continuity . equation yields the following second-order partial differential equation:
Thus both formulations are governed by the differential equation of the same type with kx == ky == I and p == q == O. However, the two formulations differ in the way the boundary conditions are imposed. To illustrate this difference, consider a typical potential flow problem to determine fluid flow around a symmetric object, as shown in Figure 5.2. Away from the object the fluid is flowing in the x direction at a known velocityzq, In the vicinity of the object the velocities change and are unknown. At a sufficient distance downstream from the object, once again we have a constant flow in the x direction. The first task is to establish the boundaries of the solution domain. On all sides of the . . . I object, we must extend the solution domain far enough so that the assumption of uniform flow in the x direction is valid (indicated by horizontal streamlines). Since the computational effort will depend on the size of the solution domain, extending the domain too far would be inefficient. However, if the domain is not large enough, the results will be inaccurate. Thus few trials may be necessary before settling on a proper size for the solution domain. It is generally recommended to extend the computational domain in all directions to about four times the size of the object. The next task is to define appropriate boundary conditions on all sides. Taking advantage of symmetry, we need to model only a quarter of the solution domain, as shown in Figure 5.3. For the stream function formulation If!, we can take any arbitrary reference value for the base because the velocities depend only on the derivatives of If!. We choose If! == 0 along x == 0 and come up with the following boundary conditions:
Boundary Conditions for Stream Function.Formulation
1. On the bottom and along the obstruction
If!==O
SELECTED APPLICATIONS OF THE 2D BVP
y ifi=Huo r
V" V'
..-
V'
.,.
V'
...
~...
.( _
.- ... ..
ifi=O
ifi=O Figure 5.3. Boundary conditions in terms of stream function for flow around an object
2. On the left side
Integrating this expression, we have
Thus 1/t varies linearly with height on this side. 3. On the top: From linear variation of 1/t on the left side with y == H, we have
4. On the right side, because of Symmetry,
81jJ == 0 8x
Note that the original form of the boundary condition 81/t/8y == Uo on the left side is not in the form that can be handled by the natural boundary condition defined for the differential equation since the normal derivative for this side is 81/t ==> -81/t ==-8n
8x
Thus we can specify 81/t/8x on the ends but not 81/t/8y. On the right side 81/t/8n == 81/t/8x, and thus we can use this boundary condition directly. . With the potential function formulation the boundary conditions on a quarter of the solution domain are as shown in Figure 5.4. On the bottom, -top, and left side we have natural boundary conditions. The right vertical symmetry line must be a potential line and therefore we must specify that 1> == constant. The actual value does not matter.
315
316
TWO·DIMENSIONAL ELEMENTS
y 8¢!8y=O
poo-.,...
f""
f
~
"...
"J'
.-
if
" "
if of
P'"
8¢!8x=uo
.-
i/>=Constant
~
8i/>/8n=O x
8¢!8y=O Figure 5.4. Boundary, conditions in termsof potentialfunction for flow aroundan object
Boundary Conditions for PotentialFunctionFormulation 1. On the top and bottom sides a¢
ay
=0
2. Along the obstruction
3. On the left end
4. On the right end
¢
= constant
5.1.2 Steady-State Heat Flow
T(x,
Consider the problem of finding the steady-state temperature distribution y) in long chimneylike structures. Assuming no temperature gradient in the longitudinal direction, we can take a unit slice of such a structure and model it as a two-dimensional problem. Using conservation of energy on a differential volume, the following governing differential equation can easily be established:
axa (aT) kx ax + aya (ky aT) ay + Q = 0 where kx and ky are thermal conductivities in the x and y directions and Q(x, y) is specified heat generation per unit volume. Typical units for k are W/m- °C or Btu/hr- ft- of and those for Q are' W/m 3 or BtuIhr· fil. The possible boundary conditions are as follows:
SELECTED APPLICATIONS OF THE 2D BVP
~
(i) Known temperature along a boundary: T(e)
= To specified
(ii) Specified heat flux along a boundary:
The sign convention for heat flow is that heat flowing into a body is positive and that out of the body is negative. Comparing with the general form this boundary condition implies a = 0 and f3 = ~qo' On an insulated boundary or across a line of symmetry there is no heat flow and thus qo = O. (iii) Convection heat loss along a boundary:
oT = h(T(e) - T ) on '"
-k-
where h is the convection coefficient, T(e) is the unknown temperature at the boundary, and Too is the temperature of the surrounding fluid. Typical units for h are W/m 2 • °C and Btn/hr -ft2. "F, Comparing with the general form, this boundary condition implies a = -h andf3 = hToo '
5.1.3
Bars Subjected to Torsion
The problem of determining stress distribution in bars of arbitrary cross section subjected to twist can be formulated in terms ?f a stress function ¢ defined as follows:
The total shear stress can be computed by taking the vector sum as follows:
All other stress components on the cross section are assumed to be zero. (See Chapter 7 for a review of basic stress and strain concepts and derivation of stress equilibrium equations.) With this assumption the stress equilibrium equations are automatically satisfied. The strain compatibility conditions lead to the following differential equation:
e
where G is the shear modulus and is the angle of twist per unit length. The boundary condition is that ¢ must be constant on the boundary. Usually ¢ = 0 is chosen on the boundary.
317
318
TWO-DIMENSIONAL ELEMENTS
The total torque is related to the stress function as follows:
Using these equations, the finite element formulation can be employed to analyze thefollowing two situations. Determining Stresses in a Bar Subjected to a Given Torque T The governing differential equation needs 0'8, the angle of twist per unit length. The torque does not show up in the governing differential equation and thus we cannot proceed directly as in other common stress analysis situations. Recognizing that the stress function ¢ depends linearly on OB, we use the following procedure:
(i) Assume arty convenient value of OBa • To clearly identify that this value is assumed, we have used the subscript a. (ii) Find the stress function solution ¢a from the governing differential equation corresponding to OBa • Compute the total torque Ta by integrating over the bar cross section using
(iii) Use ratios and proportions to determine the actual values of the angle of twist and,. ¢ as follows: ,/.
Once ¢ is known, the two nonzero stress components can be determined by simple differentiation. All other stress components are zero. Determining Torsional Constant An important use of the stress function approach is to determine the torsional constant J for a given section. The torsional constant is a property of a cross section and is related to the torque as follows:
T
= JOB
Thus, if we set OB = l , then the total torque computed is the torsional constant of the cross section:
SELECTED APPLICATIONS OF THE 20 BVP
Figure 5.5. Rectangular waveguide
(i) Set GB = 1. " (ii) Find the stress function solution ¢ by solving the governing differential equation. Compute the total torque T by integrating over the bar cross section using
T=]=2
II
¢dA
A
Recall from Chapter 4 that] values are required for space frame members. The equations given there for some simple cross sections are based on the closed-form solutions of ¢ available for those sections. For complicated sections, closed-form solutions are not possible. However, one can use the finite element formulation presented here to determine a numerical value of] for virtually any cross section.
5.1.4
Waveguides in Electromagnetics
Waveguides are hollow conductors that are employed to efficiently transmit electrical signals at microwave frequencies (in the GHz range). A waveguide with a rectangular cross section is shown in Figure 5.5. Electromagnetic waves inside a waveguide experience multiple reflections from the enclosing walls and produce discrete modes that depend on the shape and size of the waveguide, the medium within the waveguide, and the operating frequency. The two types of modes supported by a waveguide are identified as the transverse magnetic (TM) andthe transverse electric (TE). These modes can propagate only when the operating frequency is higher than a certain frequency known as the cutofffrequency, Thus determination of the cutoff frequency is a critical parameter in the design of waveguides. The governing differential equation for determination of the cutoff frequency is the Helmholtz equation, written in terms of a scalar potential if! as follows:
where lee is the unlmown cutoff frequency. The potential is related to the transverse electric field components as follows:
319
320
TWO·DIMENSIONAL ELEMENTS
For TM modes: For TE modes: where 2 0 is the characteristic wave impedance for TM modes. For the TM modes IjJ is set to 0 along the boundary. For the TE modes the normal derivative of the potential (8i/J18n) is 0 on the boundary. This differential equation is of the same form as the general BVP with kx = ky = 1, p = 12;;, and q = O. However, there is one key difference in this application, and that is the cutoff frequency kc is not known. The situation is similar to the buckling problem considered in Chapter 3. Proceeding with the standard finite element formulation in the usual manner will give rise to an eigenvalue problem. The eigenvalues of the system will correspond to the cutoff frequencies and the corresponding eigenvectors will be the wave propagation modes.
5.2
INTEGRATION BY PARTS IN HIGHER DIMENSIONS
In order to derive the weak form for two- and three-dimensional boundary value problems, we need a formula equivalent to the integration by parts for one-dimensional problems. This is accomplished by using the well-known Gauss and Green-Gauss theorems from vector calculus. These theorems are presented here in the context of a two-dimensional problem. Their extension to the three-dimensional boundary value problems is obvious. Gauss's Divergence Theorem Consider a vector of functions F The divergence of F is defined as follows:
divP
= (F1(x, y), F2(x, y)l.
= 8F1 + 8F2 8x
8y
According to Gauss's divergence theorem, the area integral of div P is equal to line integral of its dot product with the unit outer normal 11. That is,
ff
divP dA
=
i
p
T
11 de
A
The dA = dx dy is the differential area. All terms in the boundary integral must be expressed in terms of the boundary coordinate e. The de in the boundary integral is the length of a differential segment on the boundary:
de=
~d~+di
INTEGRATION BY PARTS IN HIGHERDIMENSIONS
Written explicitly,
where
11.<
and l1y are the components of the unit normal to the boundary,
Green-Gauss Theorem Given another function g(x, y), the integral of the divergence of the product gF can be written as follows:
Using the product rule, the divergence of the product gF can be written as follows:
di ( F) - a(gFI ) a(gF2 ) _ aF1 F ag aF2 F ag IV g - ax + ay - g ax + I ax + g ay + -y ? a Rearranging terms,
g ' ( F) -g _ (aF dIVg - 1+aF -2 ) + (aax ay ax Noting that the row vector in the second term is the transpose of the gradient vector of g,
ag)T ay we have
div(gF)
= g(div F) + CilglF
Thus the integral of div(gF) can be expressed as follows: .
II A
g(divF)dA +
II
C'lglFdA
=1
gFTndc
A
Rearranging terms, we get the following relationship, commonly known as the GreenGauss theorem:
Tl1dCIIg(diVF)dA= 19F A
II('i1~lFdA
321
322
TWO-DIMENSIONAL ELEMENTS
Green-Gauss Theorem as Integration by Parts in Two Dimensions The Green-Gauss theorem can be written in a form very similar to the usual one-dimensional integration by parts by taking the vector of functions with F 1 = f and' F2 = 0, i.e., F = (f(x, y), O)T. The divergence of this function is divF
= af ax
and thus applying the Green-Gauss theorem, we have
fJ agf ax
dA
.
A
Similarly, by taking the vector of functions with F 1
=0 and F2 == f, we can see that
Note that, if the derivative on function f(x, y) is with respect to x, then the boundary integral involves nx and vice versa. Similar to the one-dimensional case, this form of the Green-Gauss theorem gives an area integral in which the derivative is switched from one function to the other. We also get a boundary integral involving the product of the functions and the component of the unit normal. These forms, and their obvious extensions to three dimensions, will be used in the following section and the remaining chapters to obtain weak forms for given boundary value problems. ,/
A two-dimensional boundary value problem is stated as follows:
Example 5.1
a2 u 2 -2 ax
-
a2 u x 3 -2 + (x + y)u + - = 0; ay
u(x, 1) == sin(x);
y
u(1, y)
= sin(y);
1 < x < 3;
au
1
a/x, 2) ==x ;
au
. ax (3, y) =u(3, y)
+1
Comparing the problem with the general form, identify the terms kx ' ky , p, ... ,fJ for this problem. Classify the boundary conditions into essential and natural types. Obtain a weak form for the problem. The solution domain is a rectangle, as shown in Figure 5.6. The coefficients· in the differential equation are as follows: P =x+y;
The boundary conditions are as follows:
x y
q==-
INTEGRATION BY PARTS IN HIGHERDIMENSIONS
y
Side 3
2
Side 2 .
Side 4
Side 1
,-0----------- x 3
Figure 5.6.
Essential: On side 1, u Natural: On side 3
= sin(x); on side 4, u = sin(y).
au =;2. ay , au
au
ky ay12y = -3 ay
au
:=:}
ky ay 12y = -3;2
:=:}
a=0
[3 = -3;2
and
Natural: On side 2 au ax
= u(3,y) + 1;
au
au
kx ax 12x = 2 ax
12x = 1,
12y = O
au
:=:}
and
kx a/Ix = 2u(3, y) + 2 :=:} a = 2
[3=2
Multiplying the differential equation by the weighting function Nj , the Galerkin weighted residual is as follows:
If (
»,
x)
a2u 2-? - 3-? + (x + y)u +- NjdA Bx: ayy
= 0;
,i
= 1,2, ...
A
Using integration by parts on the first two terms,
For the rectangular domain for this example, by substituting the normal vector components, the expressions can be written more explicitly:
323
324
TWO-DIMENSIONAL ELEMENTS
First term:
2u 3 l2 a auNi(x, 1) x (0) dx + l2 2au N (3, y) x (1) dy dx dy = l3 2. 2---:xNi aX i l x=1 y=1 axx=1 aX ye l (3 au (2 au + JX=1 2 ax Ni(x, 2) x (0) dx + J 2 ax Ni(1, y) x (-l)dy y=1
-
3 l 2 2auaNi - - dx dY l x=1 ye l ax ax .
l2 au 3 l 2 a2u . 2~Nidxdy= 2- Np,y)dy ax ye l l x=1 ye I ax-
l
2 au l3l2 au aN 2- N/l,y)dy2- -a' dxdy ax xe l y=1 aX X
y=1
Second term:
2u 3 l2 - 3 aidxdy l xe l y=1 ay2N
= l3 x=1
+ +
au Ni(x, 1) x (-1)dx -3aY 2
l Fl 2 l
y=1
-3
au
x (O)dy + 7JNP,y) Y
au -3- Np,y) ay
X
l3 _1
-3
au
i(x, 2) x (l)dx 7JN Y
auaN (O)dy+ l3l2 3-a'dxdy x=1 y=1 aX X
3 l 2 -3 a2u fJu N l3 3au N - idxdy= l3 3i(x,l)dxi(x.2)dx l xe I ye l ay2N x=1 aY x=1 aY
+
3 l 2 3auaNi - - dx dY l xe l y=1 ax ax
Thus the weighted residual is
2 2au Np,y)dy- l2 2au N/l,y)dy+ l3 3au N l3 3au N i(x,l)dxi(x,2)dx aX l y=1 ax ye l _1 aY x=1 aY +
3 l 2 (-2--'+3--'+ auaN auaN ( (x+y)u+-X) N )dxdy=O ax ax ay ay y , l x=1 y=1
Consider boundary terms: First term:
l
~
au 2- Ni(3,y)dy y=1 ax
FINITE ELEMENTEQUATIONS USINGTHE GALERKIN METHOD
. au SIde 2: ax
I:I
~
. =u(3, y) + 1; Natural
2(u(3, y) + 1)Ni(3, y) dy
Second term: 2
-
au 2- N i(l , y) dy
L a= ye I
X
Side 4: u
.
sin(y); Essential
Must satisfy explicitly: -
u
(2 2 aa N i(l , y) dy
Jy=l
x
=0
Third term: 3
au
L 7JNJx,1)dx = 3
x=1
Y
Side 1: u
sin(x); Essential
Must satisfy explicitly:
(3 3 :uNi(x, 1) dx
J<=I
y
=0
Fourth term: 3
-
au
L 7JN 3
-,=1
Side 3:
i(x,2)dx
y
:~ = x3; Natural
~ 1:1 -3x3 NJx, 2) dx Thus the wealeform is
1:1
2(u(3, y) +
1)~i~3, y) dy + 1:1 -3x3 Ni(x, 2)dx
13 L2 (
+.
-,=1
ye l
auaN.1 +3auaN. -2-a1 + ( (x+y)u+ -X) N,) dxdy aX X aY Y Y
-a
=0
Admissible solutions must satisfy the EBe on sides 1 and 4.
5.3
FINITE ELEMENT EQUATIONS USING THE GALERKIN METHOD
The solution domain is discretized into elements. For simplicity of notation we do not introduceany super- or subscriptsand assume that A and C now refer to an arbitrary element area and its boundary. For the general derivation considered in this section, no specific
325
326
TWO·DIMENSIONAL ELEMENTS
shape of the element is assumed. Explicit equations for rectangular and triangular are developed in later sections. The general form of finite element equations for the problem can be written using exactly the same steps used for the one-dimensional problems. Recall from Chapter 2 that for one-dimensional problems a weak form is written by using standard steps of writing the weighted residual, integration by parts, and incorporating the natural boundary conditions. Furthermore, in the finite element form the weighting functions are the same as the interpolation functions N, used to define an assumed solution over a finite element. The same general procedure works for two-dimensional problems. The Galerkin weighted residual for the two-dimensional boundary value problem is
k ax + aya (ky aU) ay + pu + q.) N, dA = ff( axa (au) x
0;
i = 1,2•...
A
Using the Green-Gauss theorem on the first two terms in the weighted residual, we have
Thus the weighted residual is
aNi - kyau aNi + puNi+ qNi)dA = 0 Jf( kx au ax nx + kyau) a/y Nide + ff( -kx au ax & ay Iii c A. (
Splitting the boundary integral into two parts, the one over which the essential boundary condition is specified (Ce ) and the other where the natural boundary condition is specified (CII ) , and substituting the natural boundary condition, we get
As was the case for one-dimensional problems, requiring the assumed solution to satisfy the essential boundary conditions results in weighting functions that are zero over the Ce boundary, Thus with the admissible assumed solutions the weak-form equivalent to the given boundary value problem is as follows: i = 1,2•...
FINITE ELEMENT EQUATIONS USING THE GALERKIN METHOD
The assumed solution over an element is written as follows:
u(x,y) = (N1(x,y)
N
2(x,y)
...
N,1(X'Y))[~~~]
=NTd
un
811 8x
= (8NI
au 8y
= (8N I
8N2 8x
8x
8N2 8y
8y
where up u2' ••• are the unknown solutions at the element nodes. Over the boundary of the element the assumed solution must be written in terms of boundary coordinate e in the following form:
u(e)
= (NI(e)
N2(e)
...
NIl(e))[~] =N~d ull
Substituting these into the weak form, we have L
c;
(aNi(e)N~ d +[3N)de +
II
(-kx
~~iB;-d-k)'~~iB~d +pNiNTd + qNi)dA = 0;
i = 1,2, ...
II
Writing all n equations, with Ni' i = 1, 2, ... , n, together in a matrix form yields
Rearranging terms by keeping quantities involving d on the left-hand side, we get the element equations as follows:
The first two terms inside the area integral can be arranged in a more compact form by using matrices as follows:
where
ax ~J ~
and
ay
C
=
(J
~)
Thus the finite element equations are as follows:
Define n x n matrices
kk =
II
T
BCB dA;
A
Define n X 1 vectors rq =
II
qNdA;
A
Thus the element equations are
(kk + k p + lea)d =r q + rfJ The quantities lea and rfJ are a result of applied natural boundary conditions and will affect only those elements that have a boundary on which an NBC is specified. For the interior elements these terms will be zero. . In order to actually evaluate the integrals defining the element equations, we need to assume an element shape and develop appropriate interpolation functions. Any element shape can be created as long as it is useful in modeling the geometry of practical solution domains and it is possible to carry out required integrations and differentiations. Some typical elements are presented in the remaining sections and the following chapter.
RECTANGULAR FINITE ELEMENTS
5.4
RECTANGULAR FINITE ELEMENTS
A rectangular element is the simplest element that can be developed for two-dimensional problems. The assumed solutions can easily be written by using the Lagrange interpolation formula in the x and y directions. The area and boundary integrals are also straightforward to evaluate over rectangular domains. However, from a practical point of view a rectangular element is not very useful sinceit is difficult to model irregular geometries using rectangular elements alone. The rectangular elements are developed here primarily to show the procedure without integration issues complicating the discussion.
5.4.1
Four-Node Rectangular Element
A typical rectangulai element with dimensions 2a x 2b is shown in Figure 5.7. The boundary is defined by four sides of the element. The outer normals and boundary coordinate c for each side are also shown in the figure. The computations are easier if we choose a local coordinate system (s, t) with the origin at the element center. For each element these local coordinates are related to the global coordinates by a simple translation. Denoting the global coordinates of the element centroid by (xc' Yc)' we have the following relationship between the local and the global coordinates: t
=Y -
Yc
Note that with tins change of variables ds
= dx;
dt
= dy
and therefore the derivatives with respect to x and y are exactly the same as those with respect to sand t. In the local coordinate system, the nodal coordinates are (-a, -b), (a, -b), (a, b), and (-a, b). To evaluate boundary terms, for each side a boundary coordinate c is defined with the origin at the center of each side. The positive directions of these boundary coordinates are consistent with the requirement of moving counterclockwise around the boundary of the element.
y
y
n
n
n
---------- x
---:--------x
Figure 5.7. Four-node rectangular element
329
330
TWO·DIMENSIONAL ELEMENTS
Assumed Solution Assuming unknown solutions at the four corners as nodal degrees of freedom, we have a total of four degrees of freedom. The assumed finite element solutions are needed in the following form:
Over A:
U(S,
t)
= (N I (s, t)
Nz(s, t)
Over C:
Over the area the assumed solution can be written starting from a polynomial in two dimensions with four coefficients. Following the discussion in Chapter 2, we must use complete polynomials as much as possible. A complete linear polynomial in two dimensions has three terms Co + CIS + czt. For the fourth term we must choose from one of the three quadratic terms c 3st + c4s z + cst z. For an element that is appropriate for any general application the assumed solution should be symmetric; otherwise the element will give different results depending upon its orientation. Based on this consideration, we choose the st term as the fourth term, and thus a suitable polynomial solution for a rectangular element is as . follows:
The coefficients co' c 1' •.. can be expressed in terms of nodal degrees of freedom by evaluating the polynomial at the nodes as follows:
-a a 1 a 1 -a
["'] [1 U
z _ 1
u3
-
u4
-b -b -ob cI ab]['"J b ab Cz b -ab c3
Inverting the 4 x 4 matrix, we get I
4 I
-4ci
m"
I
I
4 I
4ci I
I
4
I
4
I
I
4ci. -4ci I
I
-4ij
-4ij
4ij
4ij
I 4ab
I - 4ab
I 4ab
I - 4ab
[~: J u3 u4
RECTANGULAR FINITEELEMENTS
Substituting into the polynomial, we have I
I
I
u(s, t) = (l
s
st)
t
I
4
4
4 I
I
I
4 I
-4li
4li
4li
-4li
I -4ij
I -4ij
I 4ij
I
I
I
I
- 4ab
4ab
4ab
4ij
["']
I
u2 u3
U4
- 4ab
Carrying out matrix multiplication,
(a+s)(b+tl
(a+s)(b-t)
= ( (a-s)(b-t)
U(s t) ,
----;jQb
----;jQb
4ab
N) 4
U2I] U
[ U3
= NTd
U4
N. _ (a + s)(b - t) .
(a - s)(b - t)
=
NI
4ab
;
2-
N _ (a + s)(b + t) . 34ab '
4ab'
N _ (a - s)(b 4 4ab
+ t)
where N1, •• . , N4 are the required iriterpolation functions. Note that each of the interpolation functions N, is 1 at the ith node of the element and 0 at every other node. The solutions along the elementsides are written by substituting appropriate values of sand t for each side. Expressing boundary interpolation functions in terms of e for each side, we have For side 1: For side 2: For side 3: For side 4:
a+e = e, t =- b ==* NTe,1-2 = (a-e 2Q 2Q 0 0); b-e b+e s = a, t = e ==* N{2-3 = (0 2b 2b 0); a-e ); s = -e, t = b ==* N{3-4 = (0 0 2Q a 2:c ). s = -a, t = -e ==* N{4-1 =( b2+be o 0 b-e 2b '
s
Element Equations tors: kk =
If II A
-b s; c s b -a::;; c s a -b::;; c s b
The element equations involve the following matrices and vec-
T
BeB dA-'
kp
A
r, =
-a::;; c s a
qNdA;
liJ =
=-
i
c,
II A
f3N e de
T pNN
dA;
331
332
TWO-DIMENSIONAL ELEMENTS
as ~l
i!!!:J..
as
at
c = (~
and
~
i!!!:J..
at
~)
For simplicity in integrations, it will be assumed that k", k y, p, q, a, and f3 are all constant over an element. Thus these terms can be taken out of the integrals. Substituting the assumed solution and carrying out integrations, the matrices and vectors needed for element equations can easily be written: (a+s)(b-t) 4ab